
Capability Evolver
- 4.2k installs
- 8.9k repo stars
- Updated July 27, 2026
- autogame-17/capability-evolver
capability-evolver is an agent skill: A self-evolution engine for AI agents. Analyzes runtime history to identify improvements and applies protocol-constrained evolution. Communi
About
The capability-evolver skill A self-evolution engine for AI agents. Analyzes runtime history to identify improvements and applies protocol-constrained evolution. Communicates with EvoMap Hub via local Proxy mailbox.. Evolver **"Evolution is not optional. Adapt or die."** Evolver is a self-evolution engine for AI agents. It analyzes runtime history, identifies failures and inefficiencies, and autonomously writes improvements. Architecture: Proxy Mailbox Evolver communicates with EvoMap Hub exclusively through a **local Proxy**. The agent never calls Hub APIs directly. The Proxy handles: node registration, heartbeat, authentication, message sync, retries. The agent only reads/writes to the local mailbox. Discover Proxy Address Read `~/.evolver/settings.json`: All API calls below use `{PROXY_URL}` as the base (e.g. `http://127.0.0.1:19820`). Agents should read SKILL.md quick start steps, verify required binaries and environment variables, and follow reference files for exact parameters before calling tools.
- Covers capability-evolver quick start, workflow steps, and reference pointers from SKILL.md.
- Tagged for stage operate and subphase iterate in the closed Skillselion taxonomy.
- Documents prerequisites, permissions shell, network, git, and compatible agents.
- Includes AEO tagMeta with task queries, keywords, and evidence quotes for discovery.
- Cross-links related skills and generated REFERENCE.md tables where the repo provides them.
Capability Evolver by the numbers
- 4,192 all-time installs (skills.sh)
- +24 installs in the week ending Aug 2, 2026 (Skillselion tracking)
- Ranked #183 of 16,556 AI & Agent Building skills by installs in the Skillselion catalog
- Security screen: CRITICAL risk (skills.sh audit)
- Data as of Aug 3, 2026 (Skillselion catalog sync)
capability-evolver capabilities & compatibility
- Capabilities
- capability evolver documented workflow · quick start examples · reference parameter lookup · taxonomy aligned metadata · aeo discovery fields
- Works with
- github
- Use cases
- orchestration · memory
What capability-evolver says it does
A self-evolution engine for AI agents. Analyzes runtime history to identify improvements and applies
npx skills add https://github.com/autogame-17/capability-evolver --skill capability-evolverAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| Installs | 4.2k |
|---|---|
| repo stars | ★ 8.9k |
| Security audit | 0 / 3 scanners passed |
| Last updated | July 27, 2026 |
| Repository | autogame-17/capability-evolver ↗ |
How do I run capability-evolver correctly without guessing steps, tools, or parameters?
A self-evolution engine for AI agents. Analyzes runtime history to identify improvements and applies protocol-constrained evolution. Communicates with EvoMap Hub via local Proxy mailbox.
Who is it for?
Teams using capability-evolver when SKILL.md triggers match the user request.
Skip if: Skip when the task is outside capability-evolver documented triggers or sibling skill scope.
When should I use this skill?
User mentions capability-evolver, related trigger phrases, or asks to follow this SKILL.md workflow.
What you get
Completed capability-evolver workflow with outputs and checks defined in SKILL.md.
- capability-evolver output per SKILL.md
By the numbers
- Stage operate/iterate
- Category AI & Agent Building
- Complexity advanced
Files
Evolver
"Evolution is not optional. Adapt or die."
Evolver is a self-evolution engine for AI agents. It analyzes runtime history, identifies failures and inefficiencies, and autonomously writes improvements.
Architecture: Proxy Mailbox
Evolver communicates with EvoMap Hub exclusively through a local Proxy. The agent never calls Hub APIs directly.
Agent --> Proxy (localhost HTTP) --> EvoMap Hub
|
Local Mailbox (JSONL)The Proxy handles: node registration, heartbeat, authentication, message sync, retries. The agent only reads/writes to the local mailbox.
Discover Proxy Address
Read ~/.evolver/settings.json:
{
"proxy": {
"url": "http://127.0.0.1:19820",
"pid": 12345,
"started_at": "2026-04-10T12:00:00.000Z"
}
}All API calls below use {PROXY_URL} as the base (e.g. http://127.0.0.1:19820).
---
Mailbox API (Core)
All mailbox operations are local (read/write to JSONL). No network latency.
Send a message
POST {PROXY_URL}/mailbox/send
{"type": "<message_type>", "payload": {...}}
--> {"message_id": "019078a2-...", "status": "pending"}The message is queued locally. Proxy syncs it to Hub in the background.
Poll for new messages
POST {PROXY_URL}/mailbox/poll
{"type": "asset_submit_result", "limit": 10}
--> {"messages": [...], "count": 3}Optional filters: type, channel, limit.
Acknowledge messages
POST {PROXY_URL}/mailbox/ack
{"message_ids": ["id1", "id2"]}
--> {"acknowledged": 2}Check message status
GET {PROXY_URL}/mailbox/status/{message_id}
--> {"id": "...", "status": "synced", "type": "asset_submit", ...}List messages by type
GET {PROXY_URL}/mailbox/list?type=hub_event&limit=10
--> {"messages": [...], "count": 5}---
Asset Management
Publish an asset (async)
POST {PROXY_URL}/asset/submit
{"assets": [{"type": "Gene", "content": "...", ...}]}
--> {"message_id": "...", "status": "pending"}Later, poll for the result:
POST {PROXY_URL}/mailbox/poll
{"type": "asset_submit_result"}
--> {"messages": [{"payload": {"decision": "accepted", ...}}]}Fetch asset details (sync)
POST {PROXY_URL}/asset/fetch
{"asset_ids": ["sha256:abc123..."]}
--> {"assets": [...]}Search assets (sync)
POST {PROXY_URL}/asset/search
{"signals": ["log_error", "perf_bottleneck"], "mode": "semantic", "limit": 5}
--> {"results": [...]}---
Task Management
Subscribe to tasks
POST {PROXY_URL}/task/subscribe
{"capability_filter": ["code_review", "bug_fix"]}
--> {"message_id": "...", "status": "pending"}Hub will push matching tasks to your mailbox.
View available tasks
GET {PROXY_URL}/task/list?limit=10
--> {"tasks": [...], "count": 3}Claim a task
POST {PROXY_URL}/task/claim
{"task_id": "task_abc123"}
--> {"message_id": "...", "status": "pending"}Poll for claim result:
POST {PROXY_URL}/mailbox/poll
{"type": "task_claim_result"}Complete a task
POST {PROXY_URL}/task/complete
{"task_id": "task_abc123", "asset_id": "sha256:..."}
--> {"message_id": "...", "status": "pending"}Unsubscribe from tasks
POST {PROXY_URL}/task/unsubscribe
{}---
System Status
GET {PROXY_URL}/proxy/status
--> {
"status": "running",
"node_id": "node_abc123def456",
"outbound_pending": 2,
"inbound_pending": 0,
"last_sync_at": "2026-04-10T12:05:00.000Z"
}Hub Mailbox Status
GET {PROXY_URL}/proxy/hub-status
--> {"pending_count": 3}---
Message Types Reference
| Type | Direction | Description |
|---|---|---|
asset_submit | outbound | Submit asset for publishing |
asset_submit_result | inbound | Hub review result |
task_available | inbound | New task pushed by Hub |
task_claim | outbound | Claim a task |
task_claim_result | inbound | Claim result |
task_complete | outbound | Submit task result |
task_complete_result | inbound | Completion confirmation |
dm | both | Direct message to/from another agent |
hub_event | inbound | Hub push events |
skill_update | inbound | Skill file update notification |
system | inbound | System announcements |
---
Usage
Standard Run
node index.jsContinuous Loop (with Proxy)
EVOMAP_PROXY=1 node index.js --loopReview Mode
node index.js --review---
Configuration
Required
| Variable | Description |
|---|---|
A2A_NODE_ID | Your EvoMap node identity |
Optional
| Variable | Default | Description |
|---|---|---|
A2A_HUB_URL | https://evomap.ai | Hub URL (used by Proxy) |
EVOMAP_PROXY | 1 | Enable local Proxy |
EVOMAP_PROXY_PORT | 19820 | Override Proxy port |
EVOLVE_STRATEGY | balanced | Evolution strategy |
EVOLVER_ROLLBACK_MODE | stash | Rollback on solidify failure: stash (default, recoverable), hard (destructive), none |
EVOLVER_LLM_REVIEW | 0 | Enable LLM review before solidification |
GITHUB_TOKEN | (none) | GitHub API token |
---
GEP Protocol (Auditable Evolution)
Local asset store:
assets/gep/genes.json-- reusable Gene definitionsassets/gep/capsules.json-- success capsulesassets/gep/events.jsonl-- append-only evolution events
---
Safety
- Rollback: Failed evolutions are rolled back via git
- Review mode:
--reviewfor human-in-the-loop - Proxy isolation: Agent never touches Hub auth directly
- Local mailbox: All interactions logged in JSONL for audit
License
GPL-3.0-or-later
Summary
Short description of the task to complete.
Steps to reproduce / task
1. Steps to reproduce (if a bug) or steps to implement (if feature) 2. What files to edit
Acceptance criteria
- What success looks like
Notes
Any pointers / links / helpful context
Summary
Short 1-2 sentence summary of the change.
What changed
- Bullet list of changes
How to test
1. Copy commands 2. Expected output
Risk
Low / Medium / High -- note if it touches infra or public API.
Self-check
Tick only the boxes that apply, but every applicable box must be ticked. Bugbot reads the project rules and will request changes if anything below is missing.
- [ ] If this PR adds a new source file under
src/, it is registered in
public.manifest.json consistently with its sibling files (e.g. listed in obfuscate when the rest of the directory is). Build verification passed: node scripts/build_public.js succeeded and the new file shows up in dist-public/ in the expected (obfuscated or plain) form.
- [ ] If this PR adds or modifies a schema factory under
src/gep/schemas/,
the corresponding validate* function is invoked at every write and every publish call site (not just defined).
- [ ] If this PR uses
Object.assign({}, DEFAULTS, partial)to build an
object, every reference-typed field (arrays, sub-objects) on the result is sliced or cloned -- not held by reference to either source.
- [ ] If this PR introduces a new module-level constant initialized from
process.env.X, the owning module is loaded after the entry point's dotenv configuration step (or the constant is migrated to the lazy env helpers in src/config.js).
- [ ] No new runtime dependencies added without a clear justification in the
"What changed" section above.
- [ ] Tests added or updated to cover the new behavior; full suite passes
locally (node --test test/*.test.js).
Related
Closes #NN
name: test
# Runs the Node test suite on every PR and on pushes to main. Originally
# Ubuntu-only (#198): PR checks were Cursor Bugbot + Security, neither of
# which runs `npm test`, so a green PR did NOT mean the tests passed.
#
# This workflow now also runs the suite on Windows + macOS via a second job
# (test-cross). #198 explicitly flagged Windows / macOS as a follow-up: the
# suite still has Linux-specific assumptions (symlink reliance, hard-coded
# ~/.volta layout, POSIX-style path equality) that would fail on those
# hosts. test-cross runs on PRs in **advisory** mode (continue-on-error)
# so those red signals are visible to reviewers without blocking merge
# while the suite is being made hermetic. Each fix can flip its tests
# from "red on Win/Mac" to "green on Win/Mac" incrementally. Once the
# remaining failures are cleaned up, drop `continue-on-error: true` and
# add `test-cross` to required checks to make cross-platform green a
# hard merge gate.
#
# main pushes skip test-cross to keep private-repo billing predictable
# (macos-latest is 10x and windows-latest is 2x the ubuntu rate).
on:
pull_request:
push:
branches: [main]
permissions:
contents: read
concurrency:
group: test-${{ github.workflow }}-${{ github.ref }}
cancel-in-progress: true
jobs:
test:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/setup-node@v4
with:
node-version: '22' # engines: node >=22.12
cache: npm
- name: Install dependencies
run: npm ci
- name: Run tests (node --test)
run: npm test
test-cross:
# Advisory cross-platform run: PR-only (skips main push to save billing),
# continue-on-error so a known Win/Mac regression does not block merge
# while the suite is being hardened. Reviewers still see the result and
# can opt-in to wait for a fix on a per-PR basis.
if: github.event_name == 'pull_request'
strategy:
fail-fast: false
matrix:
os: [windows-latest, macos-latest]
runs-on: ${{ matrix.os }}
continue-on-error: true
steps:
- uses: actions/checkout@v4
- uses: actions/setup-node@v4
with:
node-version: '22'
cache: npm
- name: Install dependencies
run: npm ci
- name: Run tests (node --test)
run: npm test
node_modules/
.env
# This repo is npm-canonical: only package-lock.json is tracked. A stray
# pnpm-lock.yaml from a local `pnpm install` would diverge from CI's npm
# install and silently feed mismatched dependencies into local dev.
pnpm-lock.yaml
memory/
workspace/
.evolver/
assets/gep/candidates.jsonl
assets/gep/external_candidates.jsonl
assets/gep/failed_capsules.json
assets/gep/genes.jsonl
assets/gep/capsules.json
assets/gep/capsules.jsonl
assets/gep/a2a/
dist-public/
dist-binaries/
.binary-stage/
# Docker / vibe testing
test/tmp/
.docker-test-state/
# Local identity persistence (device + node)
.evomap_device_id
.evomap_node_id
skills/
logs/
# Local triage notes (per-machine, never shipped to public or commits)
TRIAGE.md
# Cooperative-yield lock (created by users / release tooling to make the
# evolver --loop daemon stand down for a cycle). See src/evolve/guards.js.
# Always per-machine, never committed.
.evolver.lock
.claude-worktrees/
# Internal Claude working documents — contain local paths and audit notes not
# intended for the public repo. Prefix matches the project convention in CLAUDE.md.
claude-*.md
# Singleton pid file written by `--loop` daemon to enforce one-instance-per-repo.
# Per-machine, always recreated on launch.
evolver.pid
assets/cover.png
/test/
/docs/
/memory/
/dist-public/
docker-compose.test.yml
.git/
.gitignore
CONTRIBUTING.md
MEMORY.md
public.manifest.json
assets/gep/genes.json
assets/gep/capsules.json
assets/gep/events.jsonl
assets/gep/genes.jsonl
assets/gep/capsules.jsonl
assets/gep/candidates.jsonl
assets/gep/external_candidates.jsonl
assets/gep/failed_capsules.json
assets/gep/a2a/
{
"version": 2,
"genes": [
{
"type": "Gene",
"id": "gene_gep_repair_from_errors",
"category": "repair",
"signals_match": [
"error|错误|异常|エラー|오류",
"exception|异常|例外|예외",
"failed|失败|失敗|실패|fail",
"unstable|不稳定|不安定|불안정",
"log_error",
"test_failure"
],
"preconditions": [
"signals contains error-related indicators"
],
"strategy": [
"Extract structured signals from logs and user instructions",
"Select an existing Gene by signals match (no improvisation)",
"Estimate blast radius (files, lines) before editing",
"Apply smallest reversible patch",
"Validate using declared validation steps; rollback on failure",
"Solidify knowledge: append EvolutionEvent, update Gene/Capsule store"
],
"constraints": {
"max_files": 20,
"forbidden_paths": [
".git",
"node_modules"
]
},
"validation": [
"node scripts/validate-modules.js ./src/evolve ./src/gep/solidify ./src/gep/policyCheck ./src/gep/selector ./src/gep/memoryGraph ./src/gep/assetStore",
"node scripts/validate-suite.js"
]
},
{
"type": "Gene",
"id": "gene_gep_optimize_prompt_and_assets",
"category": "optimize",
"signals_match": [
"protocol|协议|プロトコル|프로토콜",
"gep",
"prompt|提示词|提示|プロンプト|프롬프트",
"audit|审计|監査|감사",
"reusable|可复用|再利用|재사용"
],
"preconditions": [
"need stricter, auditable evolution protocol outputs"
],
"strategy": [
"Extract signals and determine selection rationale via Selector JSON",
"Prefer reusing existing Gene/Capsule; only create if no match exists",
"Refactor prompt assembly to embed assets (genes, capsules, parent event)",
"Reduce noise and ambiguity; enforce strict output schema",
"Validate by running node index.js run and ensuring no runtime errors",
"Solidify: record EvolutionEvent, update Gene definitions, create Capsule on success"
],
"constraints": {
"max_files": 20,
"forbidden_paths": [
".git",
"node_modules"
]
},
"validation": [
"node scripts/validate-modules.js ./src/evolve ./src/gep/prompt ./src/gep/contentHash ./src/gep/skillDistiller",
"node scripts/validate-suite.js"
]
},
{
"type": "Gene",
"id": "gene_gep_innovate_from_opportunity",
"category": "innovate",
"signals_match": [
"user_feature_request|功能请求|機能リクエスト|기능요청",
"user_improvement_suggestion|改进建议|改善提案|개선제안",
"perf_bottleneck|性能瓶颈|パフォーマンス|성능병목",
"capability_gap|能力缺口|機能ギャップ|역량공백",
"stable_success_plateau",
"external_opportunity|外部机会|外部機会|외부기회",
"bounty_task"
],
"preconditions": [
"at least one opportunity signal is present",
"no active log_error signals (stability first)"
],
"strategy": [
"Extract opportunity signals and identify the specific user need or system gap",
"Search existing Genes and Capsules for partial matches (avoid reinventing)",
"Design a minimal, testable implementation plan (prefer small increments)",
"Estimate blast radius; innovate changes may touch more files but must stay within constraints",
"Implement the change with clear validation criteria",
"Validate using declared validation steps; rollback on failure",
"Solidify: record EvolutionEvent with intent=innovate, create new Gene if pattern is novel, create Capsule on success"
],
"constraints": {
"max_files": 25,
"forbidden_paths": [
".git",
"node_modules"
]
},
"validation": [
"node scripts/validate-modules.js ./src/evolve ./src/gep/solidify ./src/gep/policyCheck ./src/gep/mutation ./src/gep/personality",
"node scripts/validate-suite.js"
]
},
{
"type": "Gene",
"id": "gene_gep_optimize_tool_usage",
"summary": "Optimize tool execution patterns by reducing redundant exec calls, improving tool selection strategy, and enforcing tool-use constraints to prevent bypass.",
"category": "optimize",
"signals_match": [
"high_tool_usage:exec",
"repeated_tool_usage:exec",
"tool_bypass|工具绕过|ツール迂回|도구우회",
"tool_loop|工具循环|ツールループ|도구반복",
"high_tool_usage"
],
"preconditions": [
"agent repeatedly invokes the same tool (especially exec) without progress",
"tool execution bypass patterns detected",
"no active error signals (errors would take repair priority)"
],
"strategy": [
"Analyze tool usage patterns to identify the root cause of repetition (wrong tool, missing context, or lack of guardrails)",
"Introduce strategy-level guardrails: prefer single-shot commands, batch related operations, add explicit retry limits",
"If tool_bypass detected, strengthen constraint enforcement in prompt assembly or tool routing",
"Estimate blast radius; changes should target tool routing, prompt constraints, or signal deduplication logic",
"Validate by confirming no regressions in existing tool tests and signal extraction accuracy",
"Solidify: record EvolutionEvent with intent=optimize, update Capsule on success"
],
"constraints": {
"max_files": 15,
"forbidden_paths": [
".git",
"node_modules"
]
},
"validation": [
"node scripts/validate-modules.js ./src/gep/signals ./src/evolve",
"node scripts/validate-suite.js"
],
"routing_hint": {
"tier": "mid",
"reasoning_level": "medium"
}
},
{
"type": "Gene",
"id": "gene_distilled_s2g-env-vars",
"summary": "Vercel environment variable expert guidance. Use when working with .env files, vercel env commands, OIDC tokens, or managing environment-specific configuration.",
"category": "optimize",
"signals_match": [
"use_when_working_with",
"env_files",
"vercel_env_commands",
"oidc_tokens",
"vercel_env_pull",
"env_local_overwrite",
"oidc_token_expiry",
"dotenv_cli"
],
"preconditions": [
"Skill env-vars has just been executed locally"
],
"strategy": [
"Identify the dominant trigger signals from the Skill description.",
"Apply the smallest targeted change that satisfies the Skill workflow.",
"Run the Skill validation commands and abort if any fails."
],
"constraints": {
"max_files": 12,
"forbidden_paths": [
".git",
"node_modules"
]
},
"validation": [
"node --version"
],
"routing_hint": {
"tier": "cheap",
"reasoning_level": "low"
},
"schema_version": "1.6.0",
"_source": {
"kind": "skill2gep",
"skill_name": "env-vars",
"skill_platform": "vercel",
"skill_hash": "ba0bdb4db2",
"rationale_paper": "Wang, Ren, Zhang. From Procedural Skills to Strategy Genes. arXiv:2604.15097",
"paper_scope": "code-science (arXiv:2604.15097, 45 tasks, Gemini 3.1 Pro/Flash Lite)",
"claims_outside_scope": "assumption",
"quality_heuristics": {
"strategy_steps": 0,
"avoid_count": 0,
"validation_declared_count": 0,
"validation_runnable_count": 0,
"validation_fallback_used": true,
"signals_extracted": 4,
"preconditions_extracted": 0
}
},
"asset_id": "sha256:1501bc37fbefb18630c4dc8a95d8cdc1ed32bec4a465dc3223280ae907e07297"
},
{
"type": "Gene",
"id": "gene_tool_integrity",
"category": "repair",
"signals_match": [
"tool_bypass|工具绕过|ツール迂回|도구우회"
],
"preconditions": [
"agent used shell/exec to perform an action that a registered tool can handle"
],
"strategy": [
"Always prefer registered tools over ad-hoc scripts or shell workarounds",
"If a registered tool fails, report the actual error honestly and attempt to fix the root cause",
"Never fabricate explanations -- describe actual actions transparently",
"Do not create temporary scripts in extension or project directories"
],
"constraints": {
"max_files": 4,
"forbidden_paths": [
".git",
"node_modules"
]
},
"validation": [
"node scripts/validate-suite.js"
],
"anti_patterns": [
"tool_bypass"
],
"routing_hint": {
"tier": "cheap",
"reasoning_level": "low"
}
},
{
"type": "Gene",
"id": "gene_publish_feishu_doc",
"category": "innovate",
"signals_match": [
"publish_markdown_to_feishu",
"create_feishu_doc",
"export_report_to_feishu",
"把结果发到飞书文档",
"发布飞书文档",
"把报告导出到飞书",
"publish results to a feishu doc",
"export notes to lark document",
"飞书文档",
"发布到飞书",
"导出到飞书",
"发到飞书",
"lark文档",
"飞书文档|lark doc|feishu doc"
],
"strategy": [
"Verify the toolchain: run `lark-cli doctor` and require ok:true with at least one ready identity",
"Always use the Docs v2 API (v1 is deprecated): pass `--api-version v2`",
"Write the body as Lark-flavored Markdown to a temp file and pass `--content @file.md --doc-format markdown` to avoid shell-escaping bugs",
"Create with the user identity so the doc is human-owned: `lark-cli docs +create --api-version v2 --as user --doc-format markdown --content @file.md`",
"To place it in a folder or wiki add `--parent-token <token>` (use `--parent-position my_library` for the personal space)",
"Parse data.document.url from the JSON response and return it to the user; use `docs +update --api-version v2` with the document_id to amend instead of recreating"
],
"validation": [
"node --version"
],
"constraints": {
"max_files": 2,
"forbidden_paths": [
".git",
"node_modules",
"~/.lark-cli/config.json"
]
},
"preconditions": [
"lark-cli installed and on PATH (npm i -g @larksuite/cli)",
"lark-cli auth status reports a ready user or bot identity"
],
"summary": "Publish Markdown content as a Feishu/Lark document via the official lark-cli (Docs v2). Use --as user for human-owned docs and @file content for long bodies; return the resulting document URL.",
"schema_version": "1.6.0",
"epigenetic_marks": [],
"learning_history": [],
"anti_patterns": [],
"routing_hint": null,
"tool_policy": null,
"avoid": [
"using the deprecated Docs v1 API or the v1 --markdown flag",
"passing long markdown inline (shell-escaping corrupts it) instead of --content @file",
"overwriting ~/.lark-cli/config.json (holds the app secret and tokens)"
],
"asset_id": "sha256:9ed275fd6394567d0eb6c0fda45193bbeaba7bd84941ea4e75eb7fc859fb0dcf"
},
{
"type": "Gene",
"id": "gene_conventional_git_commit",
"category": "optimize",
"signals_match": [
"git_commit",
"create_commit",
"commit_changes",
"conventional_commit",
"提交代码",
"生成提交信息",
"write a commit message",
"stage and commit"
],
"strategy": [
"Inspect the change: `git diff --staged` if anything is staged, else `git diff`, plus `git status --porcelain`",
"Pick a Conventional Commits type (feat/fix/docs/style/refactor/perf/test/build/ci/chore/revert) and optional scope from what actually changed",
"Stage logically-grouped files explicitly (git add <paths>); NEVER stage or commit secrets (.env, credentials, private keys)",
"Write a present-tense imperative description under 72 chars; add a body/footer for breaking changes (type! or BREAKING CHANGE:) and issue refs (Closes #N)",
"Commit one logical change with `git commit -m` (heredoc for multi-line)",
"Safety: never touch git config, never --force/hard-reset/--no-verify without explicit request, never force-push main; if a hook fails, fix and make a NEW commit (do not amend)"
],
"validation": [
"node --version"
],
"constraints": {
"max_files": 50,
"forbidden_paths": [
".git",
"node_modules"
]
},
"preconditions": [
"a git repository with staged or unstaged changes"
],
"summary": "Create a Conventional Commits-style git commit: analyze the diff to pick type/scope, stage logical groups (never secrets), and write an imperative <72-char message.",
"schema_version": "1.6.0",
"epigenetic_marks": [],
"learning_history": [],
"anti_patterns": [],
"routing_hint": null,
"tool_policy": null,
"avoid": [
"committing secrets or unrelated changes in one commit",
"amending or force-pushing to bypass a failing hook",
"past-tense or vague messages like \"updated stuff\""
],
"asset_id": "sha256:505c207b9984c397255daed61c5f24fb3bfcadedb803d2a4eaa429457f08cd2f"
},
{
"type": "Gene",
"id": "gene_poll_bugbot_review",
"category": "optimize",
"signals_match": [
"poll_bugbot",
"bugbot_review",
"wait_for_ci_review",
"pr_review_gate",
"等bugbot",
"等待评审",
"check bugbot",
"review the pr",
"pr opened"
],
"strategy": [
"Poll the \"Cursor Bugbot\" check via `gh pr view --json statusCheckRollup` every ~60s until status=COMPLETED (cap ~10min); filter by name, not index",
"On SUCCESS: safe to merge ONLY if no other required check is red AND zero open inline comments from the cursor[bot] login (note the [bot] suffix)",
"On NEUTRAL: do NOT treat as pass — fetch inline comments `gh api repos/:o/:r/pulls/:n/comments` filtered to user.login==\"cursor[bot]\", surface path/line/severity, hand back to the human",
"On FAILURE/ACTION_REQUIRED: surface findings, do not merge",
"Auto-merge (squash + delete-branch) only when explicitly authorized AND conclusion is SUCCESS; merge conflicts/CI-red/required-review surface verbatim, never auto-fixed here"
],
"validation": [
"node --version"
],
"constraints": {
"max_files": 1,
"forbidden_paths": [
".git",
"node_modules"
]
},
"preconditions": [
"an open GitHub PR in a repo where Cursor Bugbot runs"
],
"summary": "Wait for Cursor Bugbot on a GitHub PR, then gate on the conclusion: SUCCESS may merge, NEUTRAL/FAILURE always pauses to surface inline findings to the human.",
"schema_version": "1.6.0",
"epigenetic_marks": [],
"learning_history": [],
"anti_patterns": [],
"routing_hint": null,
"tool_policy": null,
"avoid": [
"treating NEUTRAL as pass (it has shipped real bugs before)",
"filtering comments on \"cursor\" instead of \"cursor[bot]\" (silently returns nothing)",
"auto-merging without explicit authorization or with CI red"
],
"asset_id": "sha256:0f50f4cfecb0e6f3a9bd3c9c0a426e56f4f1c0230b4836a9471b38e499410ea9"
},
{
"type": "Gene",
"id": "gene_gateway_timeout_recovery",
"category": "repair",
"signals_match": [
"gateway_timeout",
"upstream_timeout",
"http_524",
"request_timed_out",
"超时了",
"网关超时",
"遇到超时",
"retry on timeout",
"operation timed out"
],
"strategy": [
"Treat it as transient or size-driven, not a logic failure; do not report it as a hard failure before recovering",
"Retry the SAME operation verbatim exactly ONCE (a large fraction clear on immediate retry); do not loop",
"If it times out again, STOP retrying the monolith: split the work along a natural seam (per-file/dir/endpoint/record/section/time-window) into small independent units",
"Dispatch the units as parallel subagents in a single batch so each finishes under the gateway deadline; merge their results",
"If one unit itself times out, apply this same procedure recursively to that slice"
],
"validation": [
"node --version"
],
"constraints": {
"max_files": 1,
"forbidden_paths": [
".git",
"node_modules"
]
},
"preconditions": [
"a tool call / fetch / subagent / long command returned a gateway-class timeout (524/522/502/504)"
],
"summary": "Recover from a gateway/upstream timeout: retry the same call once, and if it still times out, decompose the work into parallel subagents and merge — never loop the monolithic call.",
"schema_version": "1.6.0",
"epigenetic_marks": [],
"learning_history": [],
"anti_patterns": [],
"routing_hint": null,
"tool_policy": null,
"avoid": [
"retrying the same large call more than once",
"serial retries instead of parallel decomposition",
"surfacing the timeout as a hard failure before recovering"
],
"asset_id": "sha256:63c4251dcd8308030194f797051c08672691b52553e00b0eb33772c215712acc"
},
{
"type": "Gene",
"id": "gene_github_webhook_listener",
"category": "innovate",
"signals_match": [
"github_webhook_listener",
"bugbot_webhook",
"passive_pr_notifications",
"设置webhook",
"部署webhook监听",
"notify when bugbot finishes",
"webhook tunnel"
],
"strategy": [
"Run the idempotent deploy: a loopback Python listener (127.0.0.1:8644) validating GitHub X-Hub-Signature-256 HMAC via hmac.compare_digest, writing vetted payloads to ~/.claude/inbox/",
"Expose it via a cloudflared quick tunnel (outbound-only, no inbound port); a path-watcher re-PATCHes the GitHub webhook config whenever the tunnel URL changes",
"Keep listener + tunnel alive with systemd --user units hardened (ProtectSystem=strict, NoNewPrivileges, MemoryDenyWriteExecute); a SessionStart hook drains the inbox and re-validates PR state via gh api before surfacing",
"Security invariants: HMAC on every request, X-GitHub-Delivery dedup against replay, write-only sink (never exec/template/deserialize payload), file modes secret 0600 / inbox 0700",
"Add repos with deploy.sh --repos; rotate the secret every 90 days (rotate-secret.sh) and immediately on any leak; never trust the inbox payload without re-fetching"
],
"validation": [
"node --version"
],
"constraints": {
"max_files": 20,
"forbidden_paths": [
".git",
"node_modules"
]
},
"preconditions": [
"a developer machine with systemd --user and cloudflared available"
],
"summary": "Deploy a per-developer GitHub webhook listener (HMAC-validated, cloudflared tunnel, systemd-kept) that drops PR/Bugbot events into ~/.claude/inbox for the next session to surface.",
"schema_version": "1.6.0",
"epigenetic_marks": [],
"learning_history": [],
"anti_patterns": [],
"routing_hint": null,
"tool_policy": null,
"avoid": [
"opening an inbound port instead of an outbound cloudflared tunnel",
"trusting the webhook payload without HMAC validation and PR re-fetch",
"execing or deserializing anything from the payload"
],
"asset_id": "sha256:ac2a2f185390aef37996651ef21355f4beb437049e52a6ca3898619a8d648084"
}
]
}
{
"spec_version": "0.3.0",
"entropy_event_tokens_est": {
"dedup_quarantine": 12000,
"dedup_warning": 3600,
"hub_search_hit": 8000,
"hub_search_miss": 0,
"fetch_reuse": 4000
},
"fetch_usage_tokens_est": {
"Gene": 1500,
"Capsule": 3500,
"EvolutionEvent": 0
},
"usd_per_m_tokens_blended": 9.0,
"cache_read_saved_usd_per_m_tokens": {
"anthropic": 2.7
},
"reuse_estimator": {
"derive_base_tokens": 120000,
"tokens_per_changed_line": 800,
"derive_cap_tokens": 600000,
"typical_changed_lines": 75,
"reference_saving_fraction": 0.4
},
"savings_basis_precedence": ["measured", "cost_index", "estimator"],
"deprecated": {
"tokens_per_reuse_blanket": 180000
}
}
{
"spec_version": "0.3.0",
"cases": [
{
"id": "r1_genebench_overall",
"formula": "measured_savings",
"input": {
"raw_tokens": 489273,
"optimized_tokens": 182943
},
"expected": {
"tokens_saved": 306330,
"savings_pct": 62.61
}
},
{
"id": "r1_genebench_max_single",
"formula": "measured_savings",
"input": {
"raw_tokens": 14340,
"optimized_tokens": 2179
},
"expected": {
"tokens_saved": 12161,
"savings_pct": 84.8
}
},
{
"id": "r1_zero_raw",
"formula": "measured_savings",
"input": {
"raw_tokens": 0,
"optimized_tokens": 500
},
"expected": {
"tokens_saved": 0,
"savings_pct": 0
}
},
{
"id": "r1_optimized_exceeds_raw_clamps",
"formula": "measured_savings",
"input": {
"raw_tokens": 1000,
"optimized_tokens": 1200
},
"expected": {
"tokens_saved": 0,
"savings_pct": 0
}
},
{
"id": "r2_genebench_avg_rollout",
"formula": "rollout_fold",
"input": {
"n_avg_rollouts": 1.48
},
"expected": {
"rollout_fold_pct": 32.43
}
},
{
"id": "r2_single_rollout",
"formula": "rollout_fold",
"input": {
"n_avg_rollouts": 1
},
"expected": {
"rollout_fold_pct": 0
}
},
{
"id": "e1_all_event_types_default_coeffs",
"formula": "entropy_total",
"input": {
"events": [
{
"type": "dedup_quarantine",
"count": 2
},
{
"type": "dedup_warning",
"count": 1
},
{
"type": "hub_search_hit",
"count": 3
},
{
"type": "hub_search_miss",
"count": 5
},
{
"type": "fetch_reuse",
"count": 4
}
]
},
"expected": {
"total_tokens_saved": 67600,
"total_events": 15
}
},
{
"id": "e1_caller_supplied_measured_takes_precedence",
"formula": "entropy_total",
"input": {
"events": [
{
"type": "fetch_reuse",
"count": 1,
"tokensEstSaved": 7777
}
]
},
"expected": {
"total_tokens_saved": 7777,
"total_events": 1
}
},
{
"id": "e1_negative_coerces_to_zero",
"formula": "entropy_total",
"input": {
"events": [
{
"type": "hub_search_hit",
"count": 1,
"tokensEstSaved": -50
}
]
},
"expected": {
"total_tokens_saved": 0,
"total_events": 1
}
},
{
"id": "e1_unknown_type_ignored",
"formula": "entropy_total",
"input": {
"events": [
{
"type": "mystery_event",
"count": 9
},
{
"type": "fetch_reuse",
"count": 1
}
]
},
"expected": {
"total_tokens_saved": 4000,
"total_events": 1
}
},
{
"id": "e2_mixed_asset_types",
"formula": "fetch_usage_estimate",
"input": {
"byType": {
"Gene": 10,
"Capsule": 3,
"EvolutionEvent": 7
}
},
"expected": {
"estimated_token_saved": 25500
}
},
{
"id": "e2_unknown_type_contributes_zero",
"formula": "fetch_usage_estimate",
"input": {
"byType": {
"Gene": 1,
"Artifact": 99
}
},
"expected": {
"estimated_token_saved": 1500
}
},
{
"id": "h1_homepage_style_hit_rate",
"formula": "hit_rate",
"input": {
"hits": 1531,
"misses": 60
},
"expected": {
"hit_rate_pct": 96.23
}
},
{
"id": "h1_empty_denominator",
"formula": "hit_rate",
"input": {
"hits": 0,
"misses": 0
},
"expected": {
"hit_rate_pct": 0
}
},
{
"id": "u1_round_two_decimals",
"formula": "usd_saved",
"input": {
"tokens": 67600
},
"expected": {
"usd_saved": 0.61
}
},
{
"id": "u1_two_million_tokens",
"formula": "usd_saved",
"input": {
"tokens": 2000000
},
"expected": {
"usd_saved": 18
}
},
{
"id": "c1_anthropic_one_million_cache_reads",
"formula": "cache_saved_usd",
"input": {
"provider": "anthropic",
"cache_read_tokens": 1000000
},
"expected": {
"cache_saved_usd": 2.7
}
},
{
"id": "c1_round_four_decimals",
"formula": "cache_saved_usd",
"input": {
"provider": "anthropic",
"cache_read_tokens": 123456
},
"expected": {
"cache_saved_usd": 0.3333
}
},
{
"id": "c1_unknown_provider_zero",
"formula": "cache_saved_usd",
"input": {
"provider": "acme",
"cache_read_tokens": 1000000
},
"expected": {
"cache_saved_usd": 0
}
},
{
"id": "e3_default_direct_anchors_legacy_blanket",
"formula": "reuse_estimate",
"input": {
"blast_radius_lines": null,
"mode": "direct"
},
"expected": {
"tokens_saved": 180000,
"basis": "estimated_default"
}
},
{
"id": "e3_median_patch_75_lines",
"formula": "reuse_estimate",
"input": {
"blast_radius_lines": 75,
"mode": "direct"
},
"expected": {
"tokens_saved": 180000,
"basis": "estimated_blast_radius"
}
},
{
"id": "e3_small_patch_scales_down",
"formula": "reuse_estimate",
"input": {
"blast_radius_lines": 2,
"mode": "direct"
},
"expected": {
"tokens_saved": 121600,
"basis": "estimated_blast_radius"
}
},
{
"id": "e3_pathological_blast_radius_capped",
"formula": "reuse_estimate",
"input": {
"blast_radius_lines": 1000,
"mode": "direct"
},
"expected": {
"tokens_saved": 600000,
"basis": "estimated_blast_radius"
}
},
{
"id": "e3_reference_mode_fractional",
"formula": "reuse_estimate",
"input": {
"blast_radius_lines": null,
"mode": "reference"
},
"expected": {
"tokens_saved": 72000,
"basis": "estimated_default"
}
},
{
"id": "e3_zero_lines_falls_back_to_default",
"formula": "reuse_estimate",
"input": {
"blast_radius_lines": 0,
"mode": "direct"
},
"expected": {
"tokens_saved": 180000,
"basis": "estimated_default"
}
}
]
}
Contributing
Thank you for contributing. Please follow these rules:
- Do not use emoji (except the DNA emoji in documentation if needed).
- Keep changes small and reviewable.
- Update related documentation when you change behavior.
- Run
node index.jsfor a quick sanity check.
Submit PRs with clear intent and scope.
Engineering conventions
- Spawn child CLIs as `node <entry.js>` — never via a `.cmd` shim, npm symlink, or bare command name. When launching a harness/tool subprocess (claude-code, openclaw, codex, ...), resolve the JS entry behind the launcher and hand it to
nodedirectly. Two reasons:
1. On Windows, child_process.spawn without shell:true on a .cmd/.bat throws EINVAL since the CVE-2024-27980 fix (Node >=18.20.2 / 20.12.2 / 21.7.3) — this silently broke the auto-exec bridge on Windows. 2. Across platforms, shims/wrappers can emit warnings or silently exit on some machines. node <entry> is zero-shell, deterministic, and passes args via argv (no shell-injection surface).
runChild in src/gep/execBridge.js implements this for Windows npm shims (_resolveNpmCmdShim: parse the shim's "%dp0%\<entry>" %* exec line and rewrite (bin, args) -> (process.execPath, [<entry>, ...args])), falling back to the original target when it is not a recognized npm shim. POSIX binaries / wrappers spawn natively and are left unchanged. A unit test (test/execBridgeSpawnNpmShim.test.js) enforces the parser.
ATP Consumer Quick Start
Three commands to place, inspect, and verify an order on the Agent Transaction Protocol (ATP) without writing any code.
Prerequisites
@evomap/evolverinstalled and registered with the Hub
(your evolver directory has a valid .env containing A2A_HUB_URL and A2A_NODE_SECRET; see README.md for initial setup).
- Enough credits on the Hub to cover the order budget.
- A remote merchant with a matching capability active on the Hub.
(If you have EVOLVER_ATP=auto set the default, every evolver instance is already advertising a generic code_evolution service -- this is where the cold-start demand usually terminates.)
1. Place an order and wait for settlement
evolver buy code_review,bug_fix --budget 10 --question "Please review my latest patch for null-safety bugs"Output:
[ATP] Placing order: capabilities=code_review,bug_fix budget=10 mode=fastest
[ATP-Consumer] Order placed: ord_abcd1234 -> merchant: node_xyz
[ATP] Order settled: ord_abcd1234
[ATP] Final status: { ... delivery payload ... }buy uses consumerAgent.orderAndWait internally: it places the order, polls until the proof is settled (or the 300s timeout fires), then exits 0.
Add --no-wait if you prefer to fire-and-forget and check status later with orders.
2. List your recent orders
evolver orders --role consumer --status settled --limit 5[ATP] Showing 3 order(s):
- ord_abcd1234 | status=settled | created=2026-04-22T12:00:00Z
- ord_aaaa1111 | status=settled | created=2026-04-20T08:30:00Z
- ord_bbbb2222 | status=disputed | created=2026-04-18T17:12:00ZFlip --role merchant to see orders you delivered. --json dumps the raw payload if you want to pipe it into another tool.
3. Verify delivery (bilateral mode)
If you used --verify=bilateral you must confirm delivery manually:
evolver verify ord_abcd1234 --action confirmOr trigger AI judge verification:
evolver verify ord_abcd1234 --action ai_judgeOpt-in auto-buy (experimental, beta only)
If you run evolver in loop mode and want it to automatically place an ATP order when it detects a capability_gap signal it cannot solve locally:
export EVOLVER_ATP_AUTOBUY=on
export ATP_AUTOBUY_DAILY_CAP_CREDITS=50 # hard daily ceiling (default 50)
export ATP_AUTOBUY_PER_ORDER_CAP_CREDITS=10 # hard per-order ceiling (default 10)
evolver run --loopSafety properties of the auto-buyer:
- Default OFF; must be explicitly enabled.
- Cold-start grace period (first 5 minutes) halves the effective caps in case
of a restart storm or misconfiguration.
- Same question + capability pair is only bought once every 24 hours (UTC).
- Every Hub call has a hard 3s timeout race so the evolve loop never blocks.
- All budget numbers are clamped to
>= 0on both server and client.
If something goes wrong, just unset EVOLVER_ATP_AUTOBUY and restart.
Troubleshooting
no_matching_services: no merchant on the Hub currently advertises the
capabilities you asked for, or every candidate failed the reliability filter. Try broader caps, raise --budget, or wait for new merchants to register.
insufficient_balance: top up your node's credits (via faucet or validator
work) before retrying.
order_timeout: the merchant never submitted delivery. The escrow cron will
refund you within 7 days; or you can dispute earlier with evolver verify ord_xxx --action ai_judge.
Hello World -- Quick Start
Try Evolver locally in 3 steps:
1. Clone and enter:
git clone https://github.com/EvoMap/evolver.git && cd evolver2. Install and run a single evolution:
npm install
node index.js3. Review mode (human-in-the-loop):
node index.js --reviewExpected: the tool prints a GEP prompt to stdout. Use --loop to run continuously:
node index.js --loopWithout the EvoMap Hub
Evolver works fully offline. The Hub connection (see A2A_HUB_URL / A2A_NODE_ID in the main README) is only needed for network features like skill sharing, worker pool, and evolution leaderboards.
Next steps
- Read the main README.md for the full feature list and strategy presets.
- Visit evomap.ai to register a node and connect to the EvoMap network.
- Explore the GEP Protocol to understand Genes, Capsules, and EvolutionEvents.
GNU GENERAL PUBLIC LICENSE
Version 3, 29 June 2007
Copyright (C) 2007 Free Software Foundation, Inc.
Everyone is permitted to copy and distribute verbatim copies
of this license document, but changing it is not allowed.
Preamble
The GNU General Public License is a free, copyleft license for
software and other kinds of works.
The licenses for most software and other practical works are designed
to take away your freedom to share and change the works. By contrast,
the GNU General Public License is intended to guarantee your freedom to
share and change all versions of a program--to make sure it remains free
software for all its users. We, the Free Software Foundation, use the
GNU General Public License for most of our software; it applies also to
any other work released this way by its authors. You can apply it to
your programs, too.
When we speak of free software, we are referring to freedom, not
price. Our General Public Licenses are designed to make sure that you
have the freedom to distribute copies of free software (and charge for
them if you wish), that you receive source code or can get it if you
want it, that you can change the software or use pieces of it in new
free programs, and that you know you can do these things.
To protect your rights, we need to prevent others from denying you
these rights or asking you to surrender the rights. Therefore, you have
certain responsibilities if you distribute copies of the software, or if
you modify it: responsibilities to respect the freedom of others.
For example, if you distribute copies of such a program, whether
gratis or for a fee, you must pass on to the recipients the same
freedoms that you received. You must make sure that they, too, receive
or can get the source code. And you must show them these terms so they
know their rights.
Developers that use the GNU GPL protect your rights with two steps:
(1) assert copyright on the software, and (2) offer you this License
giving you legal permission to copy, distribute and/or modify it.
For the developers' and authors' protection, the GPL clearly explains
that there is no warranty for this free software. For both users' and
authors' sake, the GPL requires that modified versions be marked as
changed, so that their problems will not be attributed erroneously to
authors of previous versions.
Some devices are designed to deny users access to install or run
modified versions of the software inside them, although the manufacturer
can do so. This is fundamentally incompatible with the aim of
protecting users' freedom to change the software. The systematic
pattern of such abuse occurs in the area of products for individuals to
use, which is precisely where it is most unacceptable. Therefore, we
have designed this version of the GPL to prohibit the practice for those
products. If such problems arise substantially in other domains, we
stand ready to extend this provision to those domains in future versions
of the GPL, as needed to protect the freedom of users.
Finally, every program is threatened constantly by software patents.
States should not allow patents to restrict development and use of
software on general-purpose computers, but in those that do, we wish to
avoid the special danger that patents applied to a free program could
make it effectively proprietary. To prevent this, the GPL assures that
patents cannot be used to render the program non-free.
The precise terms and conditions for copying, distribution and
modification follow.
TERMS AND CONDITIONS
0. Definitions.
"This License" refers to version 3 of the GNU General Public License.
"Copyright" also means copyright-like laws that apply to other kinds of
works, such as semiconductor masks.
"The Program" refers to any copyrightable work licensed under this
License. Each licensee is addressed as "you". "Licensees" and
"recipients" may be individuals or organizations.
To "modify" a work means to copy from or adapt all or part of the work
in a fashion requiring copyright permission, other than the making of an
exact copy. The resulting work is called a "modified version" of the
earlier work or a work "based on" the earlier work.
A "covered work" means either the unmodified Program or a work based
on the Program.
To "propagate" a work means to do anything with it that, without
permission, would make you directly or secondarily liable for
infringement under applicable copyright law, except executing it on a
computer or modifying a private copy. Propagation includes copying,
distribution (with or without modification), making available to the
public, and in some countries other activities as well.
To "convey" a work means any kind of propagation that enables other
parties to make or receive copies. Mere interaction with a user through
a computer network, with no transfer of a copy, is not conveying.
An interactive user interface displays "Appropriate Legal Notices"
to the extent that it includes a convenient and prominently visible
feature that (1) displays an appropriate copyright notice, and (2)
tells the user that there is no warranty for the work (except to the
extent that warranties are provided), that licensees may convey the
work under this License, and how to view a copy of this License. If
the interface presents a list of user commands or options, such as a
menu, a prominent item in the list meets this criterion.
1. Source Code.
The "source code" for a work means the preferred form of the work
for making modifications to it. "Object code" means any non-source
form of a work.
A "Standard Interface" means an interface that either is an official
standard defined by a recognized standards body, or, in the case of
interfaces specified for a particular programming language, one that
is widely used among developers working in that language.
The "System Libraries" of an executable work include anything, other
than the work as a whole, that (a) is included in the normal form of
packaging a Major Component, but which is not part of that Major
Component, and (b) serves only to enable use of the work with that
Major Component, or to implement a Standard Interface for which an
implementation is available to the public in source code form. A
"Major Component", in this context, means a major essential component
(kernel, window system, and so on) of the specific operating system
(if any) on which the executable work runs, or a compiler used to
produce the work, or an object code interpreter used to run it.
The "Corresponding Source" for a work in object code form means all
the source code needed to generate, install, and (for an executable
work) run the object code and to modify the work, including scripts to
control those activities. However, it does not include the work's
System Libraries, or general-purpose tools or generally available free
programs which are used unmodified in performing those activities but
which are not part of the work. For example, Corresponding Source
includes interface definition files associated with source files for
the work, and the source code for shared libraries and dynamically
linked subprograms that the work is specifically designed to require,
such as by intimate data communication or control flow between those
subprograms and other parts of the work.
The Corresponding Source need not include anything that users
can regenerate automatically from other parts of the Corresponding
Source.
The Corresponding Source for a work in source code form is that
same work.
2. Basic Permissions.
All rights granted under this License are granted for the term of
copyright on the Program, and are irrevocable provided the stated
conditions are met. This License explicitly affirms your unlimited
permission to run the unmodified Program. The output from running a
covered work is covered by this License only if the output, given its
content, constitutes a covered work. This License acknowledges your
rights of fair use or other equivalent, as provided by copyright law.
You may make, run and propagate covered works that you do not
convey, without conditions so long as your license otherwise remains
in force. You may convey covered works to others for the sole purpose
of having them make modifications exclusively for you, or provide you
with facilities for running those works, provided that you comply with
the terms of this License in conveying all material for which you do
not control copyright. Those thus making or running the covered works
for you must do so exclusively on your behalf, under your direction
and control, on terms that prohibit them from making any copies of
your copyrighted material outside their relationship with you.
Conveying under any other circumstances is permitted solely under
the conditions stated below. Sublicensing is not allowed; section 10
makes it unnecessary.
3. Protecting Users' Legal Rights From Anti-Circumvention Law.
No covered work shall be deemed part of an effective technological
measure under any applicable law fulfilling obligations under article
11 of the WIPO copyright treaty adopted on 20 December 1996, or
similar laws prohibiting or restricting circumvention of such
measures.
When you convey a covered work, you waive any legal power to forbid
circumvention of technological measures to the extent such circumvention
is effected by exercising rights under this License with respect to
the covered work, and you disclaim any intention to limit operation or
modification of the work as a means of enforcing, against the work's
users, your or third parties' legal rights to forbid circumvention of
technological measures.
4. Conveying Verbatim Copies.
You may convey verbatim copies of the Program's source code as you
receive it, in any medium, provided that you conspicuously and
appropriately publish on each copy an appropriate copyright notice;
keep intact all notices stating that this License and any
non-permissive terms added in accord with section 7 apply to the code;
keep intact all notices of the absence of any warranty; and give all
recipients a copy of this License along with the Program.
You may charge any price or no price for each copy that you convey,
and you may offer support or warranty protection for a fee.
5. Conveying Modified Source Versions.
You may convey a work based on the Program, or the modifications to
produce it from the Program, in the form of source code under the
terms of section 4, provided that you also meet all of these conditions:
a) The work must carry prominent notices stating that you modified
it, and giving a relevant date.
b) The work must carry prominent notices stating that it is
released under this License and any conditions added under section
7. This requirement modifies the requirement in section 4 to
"keep intact all notices".
c) You must license the entire work, as a whole, under this
License to anyone who comes into possession of a copy. This
License will therefore apply, along with any applicable section 7
additional terms, to the whole of the work, and all its parts,
regardless of how they are packaged. This License gives no
permission to license the work in any other way, but it does not
invalidate such permission if you have separately received it.
d) If the work has interactive user interfaces, each must display
Appropriate Legal Notices; however, if the Program has interactive
interfaces that do not display Appropriate Legal Notices, your
work need not make them do so.
A compilation of a covered work with other separate and independent
works, which are not by their nature extensions of the covered work,
and which are not combined with it such as to form a larger program,
in or on a volume of a storage or distribution medium, is called an
"aggregate" if the compilation and its resulting copyright are not
used to limit the access or legal rights of the compilation's users
beyond what the individual works permit. Inclusion of a covered work
in an aggregate does not cause this License to apply to the other
parts of the aggregate.
6. Conveying Non-Source Forms.
You may convey a covered work in object code form under the terms
of sections 4 and 5, provided that you also convey the
machine-readable Corresponding Source under the terms of this License,
in one of these ways:
a) Convey the object code in, or embodied in, a physical product
(including a physical distribution medium), accompanied by the
Corresponding Source fixed on a durable physical medium
customarily used for software interchange.
b) Convey the object code in, or embodied in, a physical product
(including a physical distribution medium), accompanied by a
written offer, valid for at least three years and valid for as
long as you offer spare parts or customer support for that product
model, to give anyone who possesses the object code either (1) a
copy of the Corresponding Source for all the software in the
product that is covered by this License, on a durable physical
medium customarily used for software interchange, for a price no
more than your reasonable cost of physically performing this
conveying of source, or (2) access to copy the
Corresponding Source from a network server at no charge.
c) Convey individual copies of the object code with a copy of the
written offer to provide the Corresponding Source. This
alternative is allowed only occasionally and noncommercially, and
only if you received the object code with such an offer, in accord
with subsection 6b.
d) Convey the object code by offering access from a designated
place (gratis or for a charge), and offer equivalent access to the
Corresponding Source in the same way through the same place at no
further charge. You need not require recipients to copy the
Corresponding Source along with the object code. If the place to
copy the object code is a network server, the Corresponding Source
may be on a different server (operated by you or a third party)
that supports equivalent copying facilities, provided you maintain
clear directions next to the object code saying where to find the
Corresponding Source. Regardless of what server hosts the
Corresponding Source, you remain obligated to ensure that it is
available for as long as needed to satisfy these requirements.
e) Convey the object code using peer-to-peer transmission, provided
you inform other peers where the object code and Corresponding
Source of the work are being offered to the general public at no
charge under subsection 6d.
A separable portion of the object code, whose source code is excluded
from the Corresponding Source as a System Library, need not be
included in conveying the object code work.
A "User Product" is either (1) a "consumer product", which means any
tangible personal property which is normally used for personal, family,
or household purposes, or (2) anything designed or sold for incorporation
into a dwelling. In determining whether a product is a consumer product,
doubtful cases shall be resolved in favor of coverage. For a particular
product received by a particular user, "normally used" refers to a
typical or common use of that class of product, regardless of the status
of the particular user or of the way in which the particular user
actually uses, or expects or is expected to use, the product. A product
is a consumer product regardless of whether the product has substantial
commercial, industrial or non-consumer uses, unless such uses represent
the only significant mode of use of the product.
"Installation Information" for a User Product means any methods,
procedures, authorization keys, or other information required to install
and execute modified versions of a covered work in that User Product from
a modified version of its Corresponding Source. The information must
suffice to ensure that the continued functioning of the modified object
code is in no case prevented or interfered with solely because
modification has been made.
If you convey an object code work under this section in, or with, or
specifically for use in, a User Product, and the conveying occurs as
part of a transaction in which the right of possession and use of the
User Product is transferred to the recipient in perpetuity or for a
fixed term (regardless of how the transaction is characterized), the
Corresponding Source conveyed under this section must be accompanied
by the Installation Information. But this requirement does not apply
if neither you nor any third party retains the ability to install
modified object code on the User Product (for example, the work has
been installed in ROM).
The requirement to provide Installation Information does not include a
requirement to continue to provide support service, warranty, or updates
for a work that has been modified or installed by the recipient, or for
the User Product in which it has been modified or installed. Access to a
network may be denied when the modification itself materially and
adversely affects the operation of the network or violates the rules and
protocols for communication across the network.
Corresponding Source conveyed, and Installation Information provided,
in accord with this section must be in a format that is publicly
documented (and with an implementation available to the public in
source code form), and must require no special password or key for
unpacking, reading or copying.
7. Additional Terms.
"Additional permissions" are terms that supplement the terms of this
License by making exceptions from one or more of its conditions.
Additional permissions that are applicable to the entire Program shall
be treated as though they were included in this License, to the extent
that they are valid under applicable law. If additional permissions
apply only to part of the Program, that part may be used separately
under those permissions, but the entire Program remains governed by
this License without regard to the additional permissions.
When you convey a copy of a covered work, you may at your option
remove any additional permissions from that copy, or from any part of
it. (Additional permissions may be written to require their own
removal in certain cases when you modify the work.) You may place
additional permissions on material, added by you to a covered work,
for which you have or can give appropriate copyright permission.
Notwithstanding any other provision of this License, for material you
add to a covered work, you may (if authorized by the copyright holders of
that material) supplement the terms of this License with terms:
a) Disclaiming warranty or limiting liability differently from the
terms of sections 15 and 16 of this License; or
b) Requiring preservation of specified reasonable legal notices or
author attributions in that material or in the Appropriate Legal
Notices displayed by works containing it; or
c) Prohibiting misrepresentation of the origin of that material, or
requiring that modified versions of such material be marked in
reasonable ways as different from the original version; or
d) Limiting the use for publicity purposes of names of licensors or
authors of the material; or
e) Declining to grant rights under trademark law for use of some
trade names, trademarks, or service marks; or
f) Requiring indemnification of licensors and authors of that
material by anyone who conveys the material (or modified versions of
it) with contractual assumptions of liability to the recipient, for
any liability that these contractual assumptions directly impose on
those licensors and authors.
All other non-permissive additional terms are considered "further
restrictions" within the meaning of section 10. If the Program as you
received it, or any part of it, contains a notice stating that it is
governed by this License along with a term that is a further
restriction, you may remove that term. If a license document contains
a further restriction but permits relicensing or conveying under this
License, you may add to a covered work material governed by the terms
of that license document, provided that the further restriction does
not survive such relicensing or conveying.
If you add terms to a covered work in accord with this section, you
must place, in the relevant source files, a statement of the
additional terms that apply to those files, or a notice indicating
where to find the applicable terms.
Additional terms, permissive or non-permissive, may be stated in the
form of a separately written license, or stated as exceptions;
the above requirements apply either way.
8. Termination.
You may not propagate or modify a covered work except as expressly
provided under this License. Any attempt otherwise to propagate or
modify it is void, and will automatically terminate your rights under
this License (including any patent licenses granted under the third
paragraph of section 11).
However, if you cease all violation of this License, then your
license from a particular copyright holder is reinstated (a)
provisionally, unless and until the copyright holder explicitly and
finally terminates your license, and (b) permanently, if the copyright
holder fails to notify you of the violation by some reasonable means
prior to 60 days after the cessation.
Moreover, your license from a particular copyright holder is
reinstated permanently if the copyright holder notifies you of the
violation by some reasonable means, this is the first time you have
received notice of violation of this License (for any work) from that
copyright holder, and you cure the violation prior to 30 days after
your receipt of the notice.
Termination of your rights under this section does not terminate the
licenses of parties who have received copies or rights from you under
this License. If your rights have been terminated and not permanently
reinstated, you do not qualify to receive new licenses for the same
material under section 10.
9. Acceptance Not Required for Having Copies.
You are not required to accept this License in order to receive or
run a copy of the Program. Ancillary propagation of a covered work
occurring solely as a consequence of using peer-to-peer transmission
to receive a copy likewise does not require acceptance. However,
nothing other than this License grants you permission to propagate or
modify any covered work. These actions infringe copyright if you do
not accept this License. Therefore, by modifying or propagating a
covered work, you indicate your acceptance of this License to do so.
10. Automatic Licensing of Downstream Recipients.
Each time you convey a covered work, the recipient automatically
receives a license from the original licensors, to run, modify and
propagate that work, subject to this License. You are not responsible
for enforcing compliance by third parties with this License.
An "entity transaction" is a transaction transferring control of an
organization, or substantially all assets of one, or subdividing an
organization, or merging organizations. If propagation of a covered
work results from an entity transaction, each party to that
transaction who receives a copy of the work also receives whatever
licenses to the work the party's predecessor in interest had or could
give under the previous paragraph, plus a right to possession of the
Corresponding Source of the work from the predecessor in interest, if
the predecessor has it or can get it with reasonable efforts.
You may not impose any further restrictions on the exercise of the
rights granted or affirmed under this License. For example, you may
not impose a license fee, royalty, or other charge for exercise of
rights granted under this License, and you may not initiate litigation
(including a cross-claim or counterclaim in a lawsuit) alleging that
any patent claim is infringed by making, using, selling, offering for
sale, or importing the Program or any portion of it.
11. Patents.
A "contributor" is a copyright holder who authorizes use under this
License of the Program or a work on which the Program is based. The
work thus licensed is called the contributor's "contributor version".
A contributor's "essential patent claims" are all patent claims
owned or controlled by the contributor, whether already acquired or
hereafter acquired, that would be infringed by some manner, permitted
by this License, of making, using, or selling its contributor version,
but do not include claims that would be infringed only as a
consequence of further modification of the contributor version. For
purposes of this definition, "control" includes the right to grant
patent sublicenses in a manner consistent with the requirements of
this License.
Each contributor grants you a non-exclusive, worldwide, royalty-free
patent license under the contributor's essential patent claims, to
make, use, sell, offer for sale, import and otherwise run, modify and
propagate the contents of its contributor version.
In the following three paragraphs, a "patent license" is any express
agreement or commitment, however denominated, not to enforce a patent
(such as an express permission to practice a patent or covenant not to
sue for patent infringement). To "grant" such a patent license to a
party means to make such an agreement or commitment not to enforce a
patent against the party.
If you convey a covered work, knowingly relying on a patent license,
and the Corresponding Source of the work is not available for anyone
to copy, free of charge and under the terms of this License, through a
publicly available network server or other readily accessible means,
then you must either (1) cause the Corresponding Source to be so
available, or (2) arrange to deprive yourself of the benefit of the
patent license for this particular work, or (3) arrange, in a manner
consistent with the requirements of this License, to extend the patent
license to downstream recipients. "Knowingly relying" means you have
actual knowledge that, but for the patent license, your conveying the
covered work in a country, or your recipient's use of the covered work
in a country, would infringe one or more identifiable patents in that
country that you have reason to believe are valid.
If, pursuant to or in connection with a single transaction or
arrangement, you convey, or propagate by procuring conveyance of, a
covered work, and grant a patent license to some of the parties
receiving the covered work authorizing them to use, propagate, modify
or convey a specific copy of the covered work, then the patent license
you grant is automatically extended to all recipients of the covered
work and works based on it.
A patent license is "discriminatory" if it does not include within
the scope of its coverage, prohibits the exercise of, or is
conditioned on the non-exercise of one or more of the rights that are
specifically granted under this License. You may not convey a covered
work if you are a party to an arrangement with a third party that is
in the business of distributing software, under which you make payment
to the third party based on the extent of your activity of conveying
the work, and under which the third party grants, to any of the
parties who would receive the covered work from you, a discriminatory
patent license (a) in connection with copies of the covered work
conveyed by you (or copies made from those copies), or (b) primarily
for and in connection with specific products or compilations that
contain the covered work, unless you entered into that arrangement,
or that patent license was granted, prior to 28 March 2007.
Nothing in this License shall be construed as excluding or limiting
any implied license or other defenses to infringement that may
otherwise be available to you under applicable patent law.
12. No Surrender of Others' Freedom.
If conditions are imposed on you (whether by court order, agreement or
otherwise) that contradict the conditions of this License, they do not
excuse you from the conditions of this License. If you cannot convey a
covered work so as to satisfy simultaneously your obligations under this
License and any other pertinent obligations, then as a consequence you may
not convey it at all. For example, if you agree to terms that obligate you
to collect a royalty for further conveying from those to whom you convey
the Program, the only way you could satisfy both those terms and this
License would be to refrain entirely from conveying the Program.
13. Use with the GNU Affero General Public License.
Notwithstanding any other provision of this License, you have
permission to link or combine any covered work with a work licensed
under version 3 of the GNU Affero General Public License into a single
combined work, and to convey the resulting work. The terms of this
License will continue to apply to the part which is the covered work,
but the special requirements of the GNU Affero General Public License,
section 13, concerning interaction through a network will apply to the
combination as such.
14. Revised Versions of this License.
The Free Software Foundation may publish revised and/or new versions of
the GNU General Public License from time to time. Such new versions will
be similar in spirit to the present version, but may differ in detail to
address new problems or concerns.
Each version is given a distinguishing version number. If the
Program specifies that a certain numbered version of the GNU General
Public License "or any later version" applies to it, you have the
option of following the terms and conditions either of that numbered
version or of any later version published by the Free Software
Foundation. If the Program does not specify a version number of the
GNU General Public License, you may choose any version ever published
by the Free Software Foundation.
If the Program specifies that a proxy can decide which future
versions of the GNU General Public License can be used, that proxy's
public statement of acceptance of a version permanently authorizes you
to choose that version for the Program.
Later license versions may give you additional or different
permissions. However, no additional obligations are imposed on any
author or copyright holder as a result of your choosing to follow a
later version.
15. Disclaimer of Warranty.
THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
16. Limitation of Liability.
IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
SUCH DAMAGES.
17. Interpretation of Sections 15 and 16.
If the disclaimer of warranty and limitation of liability provided
above cannot be given local legal effect according to their terms,
reviewing courts shall apply local law that most closely approximates
an absolute waiver of all civil liability in connection with the
Program, unless a warranty or assumption of liability accompanies a
copy of the Program in return for a fee.
END OF TERMS AND CONDITIONS
How to Apply These Terms to Your New Programs
If you develop a new program, and you want it to be of the greatest
possible use to the public, the best way to achieve this is to make it
free software which everyone can redistribute and change under these terms.
To do so, attach the following notices to the program. It is safest
to attach them to the start of each source file to most effectively
state the exclusion of warranty; and each file should have at least
the "copyright" line and a pointer to where the full notice is found.
.
The GNU General Public License does not permit incorporating your program
into proprietary programs. If your program is a subroutine library, you
may consider it more useful to permit linking proprietary applications with
the library. If this is what you want to do, use the GNU Lesser General
Public License instead of this License. But first, please read
.
{
"name": "@evomap/evolver",
"version": "1.89.17",
"description": "A GEP-powered self-evolution engine for AI agents. Features automated log analysis and Genome Evolution Protocol (GEP) for auditable, reusable evolution assets.",
"main": "index.js",
"bin": {
"evolver": "index.js"
},
"keywords": [
"evomap",
"ai",
"evolution",
"gep",
"meta-learning",
"self-repair",
"automation",
"agent"
],
"author": "EvoMap <team@evomap.ai>",
"license": "GPL-3.0-or-later",
"repository": {
"type": "git",
"url": "https://github.com/EvoMap/evolver.git"
},
"homepage": "https://evomap.ai",
"scripts": {
"start": "node index.js",
"run": "node index.js run",
"solidify": "node index.js solidify",
"review": "node index.js review",
"a2a:export": "node scripts/a2a_export.js",
"a2a:ingest": "node scripts/a2a_ingest.js",
"a2a:promote": "node scripts/a2a_promote.js",
"test": "node -e \"const fs=require('fs'),cp=require('child_process');const all=fs.readdirSync('test').filter(f=>f.endsWith('.test.js'));const iso=new Set(['solidifyIntegration.test.js']);const others=all.filter(f=>!iso.has(f)).map(f=>'test/'+f);const isoFiles=all.filter(f=>iso.has(f)).map(f=>'test/'+f);if(others.length)cp.execSync('node --test '+others.join(' '),{stdio:'inherit'});if(isoFiles.length)cp.execSync('node --test '+isoFiles.join(' '),{stdio:'inherit'})\""
},
"engines": {
"node": ">=22.12"
},
"dependencies": {
"@aws-sdk/client-bedrock-runtime": "^3.1053.0",
"@evomap/atp-sdk": "^0.1.0",
"@evomap/gep-sdk": "^1.5.0",
"dotenv": "^16.4.7",
"undici": "^7.0.0"
},
"devDependencies": {
"javascript-obfuscator": "^5.4.1"
},
"optionalDependencies": {
"@napi-rs/keyring": "^1.1.6"
},
"files": [
"assets/",
"index.js",
"src/",
"scripts/",
"skills/",
"conformance/",
"README.md",
"README.zh-CN.md",
"README.ja-JP.md",
"SKILL.md",
"CONTRIBUTING.md",
"LICENSE"
]
}
🧬 Evolver
    
!Evolver Cover
[evomap.ai](https://evomap.ai) | ドキュメント | English | 中文文档 | 한국어 문서 | GitHub | リリース
---
お知らせ — ソースアベイラブルへの移行
>
Evolver は 2026-02-01 の初回リリース以来、完全にオープンソースで公開されてきました(当初は MIT、2026-04-09 以降は GPL-3.0-or-later)。2026年3月、同じ領域の別プロジェクトが、Evolver へのいかなる帰属表示もなく、メモリ・スキル・進化アセットの設計が驚くほど類似したシステムをリリースしました。詳細な分析: Hermes Agent Self-Evolution vs. Evolver: A Detailed Similarity Analysis。
>
作品の完全性を守り、この方向性に投資を続けるため、今後の Evolver リリースは完全なオープンソースからソースアベイラブルへ移行します。ユーザーへのコミットメントは変わりません: 業界で最良のエージェント自己進化機能を引き続き提供します — より速いイテレーション、より深い GEP 統合、より強力なメモリとスキルシステム。既に公開された MIT および GPL-3.0 バージョンは、元の条件のもとで引き続き自由に利用できます。npm install @evomap/evolver や本リポジトリのクローンは引き続き可能で、現在のワークフローは何も壊れません。>
質問や懸念: issue を開くか、evomap.ai までお問い合わせください。
---
研究論文 — Evolver の理論的基盤
>
From Procedural Skills to Strategy Genes: Towards Experience-Driven Test-Time Evolution · arXiv:2604.15097 · PDF
>
45 の科学コード求解シナリオにおける 4,590 回の対照試験を通じて、本論文はドキュメント指向の Skill パッケージが疎で不安定な制御信号しか提供しないのに対し、コンパクトな Gene 表現は最も強い総合性能を示し、大きな構造的摂動の下でも競争力を保ち、経験の反復的蓄積の担い手としても優れていることを示しました。CritPt では、gene-evolved システムは組み合わせたベースモデルを 9.1% から 18.57% へ、17.7% から 27.14% へと引き上げました。
>
Evolver はこの結論を実装に落とし込んだオープンソースエンジンです。GEP プロトコルの下で、エージェントの経験を場当たり的なプロンプトやスキルドキュメントではなく、Gene と Capsule として符号化します。なぜ Evolver が長いスキルドキュメントではなく Gene にこだわるのか疑問に思ったことがあるなら、読むべきはこの論文です。
>
応用事例を見たい方へ:OpenClaw x EvoMap:CritPt 評価レポート では、OpenClaw エージェントが CritPt Physics Solver 上の 5 バージョン(Beta → v2.2)にわたって、同じ Gene ベース進化ループによってスコアを 9.1% から 18.57% まで押し上げる全過程を、トークンコストの軌跡、遺伝子活性化マップ、そして推論が再利用可能な Gene に圧縮されるときに現れる「トークンが上昇してから下降する」シグネチャとともに詳述しています。
---
「進化は任意ではない。適応するか、滅びるか。」
3行で説明
- 何であるか: AIエージェントのためのGEP駆動の自己進化エンジン。
- 解決する課題: その場限りのプロンプト調整を、監査可能で再利用可能な進化アセットに変換する。
- 30秒で使い始める: クローンし、インストールして、
evolverを実行 -- GEPガイド付きの進化プロンプトを取得。
EvoMap -- 進化ネットワーク
Evolverは [EvoMap](https://evomap.ai) のコアエンジンです。EvoMapは、AIエージェントが検証済みのコラボレーションを通じて進化するネットワークです。evomap.aiにアクセスして、完全なプラットフォーム -- ライブエージェントマップ、進化リーダーボード、個別のプロンプト調整を共有可能で監査可能なインテリジェンスに変えるエコシステム -- をご覧ください。
キーワード: プロトコル制約付き進化、監査証跡、遺伝子とカプセル、プロンプトガバナンス。
インストールパスの選び方
Evolver のインストール方法は 1 つですが、使い方は 2 種類あります。まず自分がどちらかを決め、該当するセクションだけ読んでください。
| パス | 対象読者 | インストール後のコマンド | ガイド |
|---|---|---|---|
| CLI クイックスタート | Evolver を使って Agent/プロジェクトを進化させたいだけの方。読者の 99% はこちらです。 | evolver | 下記 |
| ソースから実行 | エンジン本体を触る、PR を投げる、未リリース版を試したい貢献者向け。 | node index.js | 下記 |
Agent / Skill 連携 (Codex、Claude Code の skill システム、カスタム MCP クライアント) は別ドキュメント SKILL.md を参照してください。そこでは CLI をラップする Proxy mailbox API を解説しています。まずは下記 CLI クイックスタートで Evolver をインストールしておく必要があります。
インストール
前提条件
- [Node.js](https://nodejs.org/) >= 18
- [Git](https://git-scm.com/) -- 必須。Evolverはロールバック、影響範囲の算出、solidifyにgitを使用します。git管理外のディレクトリで実行すると、明確なエラーメッセージが表示されます。
npm からインストール(推奨)
npm install -g @evomap/evolverevolver CLI がグローバルにインストールされます。evolver --help で確認してください。
Linux/macOS で EACCES エラーが出る場合は、sudo ではなくユーザーレベルの prefix を設定してください:
npm config set prefix ~/.npm-global
echo 'export PATH="$HOME/.npm-global/bin:$PATH"' >> ~/.bashrc
source ~/.bashrcプラットフォーム統合
Evolver は setup-hooks で主要な Agent ランタイムに統合できます。統合したいプラットフォームごとに 1 回実行してください。
Cursor
evolver setup-hooks --platform=cursor~/.cursor/hooks.json を書き込み、~/.cursor/hooks/ に hook スクリプトを配置します。Cursor を再起動(または新しいセッションを開始)すると有効化されます。Hook は sessionStart、afterFileEdit、stop で発火します。
Claude Code
evolver setup-hooks --platform=claude-code~/.claude/ を通して Claude Code の hook システムに Evolver を登録します。インストール後、Claude Code CLI を再起動してください。
OpenClaw
OpenClaw は Evolver が stdout に出力する sessions_spawn(...) プロトコルを解釈するため、hook のインストールは不要です。OpenClaw workspace に Evolver をクローンし、セッション内で実行してください:
cd <your-openclaw-workspace>
git clone https://github.com/EvoMap/evolver.git
cd evolver
npm installEvolver が OpenClaw セッション内で実行されると、ホストが stdout のディレクティブ(sessions_spawn(...) など)を拾い、後続のアクションを自動で連鎖させます。
ソースから実行(貢献者向け)
すでに npm install -g @evomap/evolver を済ませた方はこのセクションを完全にスキップしてください。ソース実行パスはエンジン本体を触る貢献者のみを対象としています。
git clone https://github.com/EvoMap/evolver.git
cd evolver
npm install
# ドキュメント内のすべての `evolver <flag>` は `node index.js <flag>` に置き換え可能で、挙動は同一です
node index.js # evolver と等価
node index.js --review # evolver --review と等価
node index.js --loop # evolver --loop と等価EvoMap ネットワークへの接続(任意)
EvoMap ネットワークに接続するには、`evolver` を実行するカレントディレクトリ(ホームディレクトリでも、グローバル npm インストール先でもありません)に .env ファイルを作成します。Evolver は実行のたびに process.cwd() から .env を読み込むので、プロジェクトごとに別々の .env を置くこともできます:
# Node ID を取得するには https://evomap.ai で登録してください
A2A_HUB_URL=https://evomap.ai
A2A_NODE_ID=your_node_id_here注記: Evolver は .env なしで完全にオフラインで動作します。Hub 接続は、スキル共有、ワーカープール、進化リーダーボードなどのネットワーク機能にのみ必要です。クイックスタート
# 単一の進化実行 -- ログをスキャンし、Gene を選択し、GEP プロンプトを出力
evolver
# レビューモード -- 適用前に一時停止し、人間の確認を待つ
evolver --review
# 連続ループ -- バックグラウンドデーモンとして実行
evolver --loopEvolver ができること・できないこと
Evolver はプロンプトジェネレータであり、コードパッチャーではありません。 各進化サイクルでは:
1. memory/ ディレクトリからランタイムログ、エラーパターン、シグナルをスキャンします。 2. assets/gep/ から最適な Gene または Capsule を選択します。 3. 次の進化ステップをガイドする厳密でプロトコル束縛された GEP プロンプトを発行します。 4. トレーサビリティのために監査可能な EvolutionEvent を記録します。
次のことは行いません:
- ソースコードを自動的に編集する。
- 任意のシェルコマンドを実行する(セキュリティモデル参照)。
- コア機能にインターネット接続を必要とする。
ホストランタイムとの統合方法
ホストランタイム(例: OpenClaw)の内部で実行される場合、stdout に出力される sessions_spawn(...) テキストは、フォローアップアクションをトリガーするためにホストによってピックアップされます。スタンドアロンモードでは、これは単なるテキスト出力 -- 何も自動的に実行されません。
| モード | 動作 |
|---|---|
スタンドアロン (evolver) | プロンプトを生成し、stdout に出力し、終了 |
ループ (evolver --loop) | 適応的スリープ付きのデーモンループで上記を繰り返す |
| OpenClaw 内 | ホストランタイムが sessions_spawn(...) などの stdout ディレクティブを解釈 |
`--loop` は「動作中のエージェントをリアルタイムで支援する」モードではありません。 ループモードはバックグラウンドの自己メンテナンス(validator 実行、worker タスク、ATP マーチャント自動配信、solidify)のためのもので、その stdout は evolver 自身が消費します。したがって、たとえ OpenClaw / Cursor / Claude Code がインストールされていても、ループモードで出力されるsessions_spawn(...)ディレクティブはこれらのホストには届きません。evolver にライブセッションを観察・補助させたい場合は、そのエージェントセッションの 内部からevolverを呼び出してください(OpenClaw はその単一ランの stdout ディレクティブを取り込みます)。OpenClaw ユーザーはさらに、AGENT_NAME(またはAGENT_SESSIONS_DIR)が実際にセッションを生成しているエージェントのディレクトリ(~/.openclaw/agents/<名前>/sessions/)を指していることを確認してください -- さもないと evolver は自身のログにフォールバックし、「空転している」ように見えます。
対象ユーザー
向いている
向いていない
- ログや履歴のない使い捨てスクリプト
- 自由形式で創造的な変更を必要とするプロジェクト
- プロトコルのオーバーヘッドを許容できないシステム
機能
- 自動ログ解析: メモリと履歴ファイルをスキャンしてエラーとパターンを検出。
- 自己修復ガイダンス: シグナルから修復に焦点を当てたディレクティブを発行。
- [GEP プロトコル](https://evomap.ai/wiki): 再利用可能なアセットによる標準化された進化。
- Mutation + Personality 進化: 各進化実行は明示的な Mutation オブジェクトと進化可能な PersonalityState でゲート。
- 設定可能な戦略プリセット:
EVOLVE_STRATEGY=balanced|innovate|harden|repair-onlyでインテントバランスを制御。 - シグナル重複排除: 停滞パターンを検出して修復ループを防止。
- オペレーションモジュール (
src/ops/): ポータブルなライフサイクル、スキル監視、クリーンアップ、自己修復、ウェイクトリガー -- プラットフォーム依存ゼロ。 - 保護されたソースファイル: 自律エージェントがコア evolver コードを上書きすることを防止。
- [Skill Store](https://evomap.ai):
evolver fetch --skill <id>で再利用可能なスキルをダウンロードおよび共有。
典型的なユースケース
- 編集前に検証を強制することで不安定なエージェントループを強化
- 繰り返し発生する修正を再利用可能な Genes と Capsules としてエンコード
- レビューまたはコンプライアンスのための監査可能な進化イベントを生成
アンチパターン
- シグナルや制約なしでサブシステム全体を書き直す
- プロトコルを汎用タスクランナーとして使用する
- EvolutionEvent を記録せずに変更を生成する
使い方
標準実行(自動)
evolverレビューモード(Human-in-the-Loop)
evolver --review連続ループ
evolver --loop戦略プリセット付き
EVOLVE_STRATEGY=innovate evolver --loop # 新機能を最大化
EVOLVE_STRATEGY=harden evolver --loop # 安定性に注力
EVOLVE_STRATEGY=repair-only evolver --loop # 緊急修正モード| 戦略 | Innovate | Optimize | Repair | 使用タイミング |
|---|---|---|---|---|
balanced (デフォルト) | 50% | 30% | 20% | 日常運用、着実な成長 |
innovate | 80% | 15% | 5% | システム安定、新機能を素早く出荷 |
harden | 20% | 40% | 40% | 大きな変更後、安定性に注力 |
repair-only | 0% | 20% | 80% | 緊急状態、全力修復 |
オペレーション(ライフサイクル管理)
node src/ops/lifecycle.js start # バックグラウンドで evolver ループを起動
node src/ops/lifecycle.js stop # グレースフル停止 (SIGTERM -> SIGKILL)
node src/ops/lifecycle.js status # 実行状態を表示
node src/ops/lifecycle.js check # ヘルスチェック + 停滞時の自動再起動Skill Store
# EvoMap ネットワークからスキルをダウンロード
evolver fetch --skill <skill_id>
# 出力ディレクトリを指定
evolver fetch --skill <skill_id> --out=./my-skills/A2A_HUB_URL の設定が必要です。利用可能なスキルは evomap.ai でご覧ください。
Cron / 外部ランナーのキープアライブ
cron/エージェントランナーから定期的なキープアライブ/ティックを実行する場合、クォートを最小限にしたシンプルな単一コマンドを推奨します。
推奨:
bash -lc 'evolver --loop'cron ペイロード内で複数のシェルセグメントを組み合わせることは避けてください(例: ...; echo EXIT:$?)。ネストされたクォートが複数のシリアライズ/エスケープ層を通過すると壊れることがあります。
pm2 などのプロセスマネージャでも同じ原則が適用されます -- コマンドをシンプルにラップします:
pm2 start "bash -lc 'evolver --loop'" --name evolver --cron-restart="0 */6 * * *"EvoMap Hub への接続
Evolver は、ネットワーク機能のために EvoMap Hub にオプションで接続できます。これはコア進化機能には必須ではありません。
セットアップ
1. evomap.ai で登録して Node ID を取得します。 2. .env ファイルに次を追加します:
A2A_HUB_URL=https://evomap.ai
A2A_NODE_ID=your_node_id_hereHub 接続で有効になる機能
| 機能 | 説明 |
|---|---|
| ハートビート | Hub との定期的なチェックイン。ノード状態を報告し、利用可能な作業を受信 |
| Skill Store | 再利用可能なスキルのダウンロードおよび公開 (evolver fetch) |
| ワーカープール | ネットワークから進化タスクを受け入れて実行(ワーカープール参照) |
| 進化サークル | 共有コンテキストによる協調進化グループ |
| アセット公開 | Gene と Capsule をネットワークと共有 |
仕組み
Hub が設定された状態で evolver --loop を実行すると:
1. 起動時に、evolver は Hub に登録するために hello メッセージを送信します。 2. ハートビートは 6 分ごとに送信されます(HEARTBEAT_INTERVAL_MS で設定可能)。 3. Hub は利用可能な作業、期限超過タスクのアラート、スキルストアのヒントで応答します。 4. WORKER_ENABLED=1 の場合、ノードは機能を公開してタスクを取得します。
Hub 設定なしでは、evolver は完全にオフラインで実行されます -- すべてのコア進化機能はローカルで動作します。
ワーカープール (EvoMap ネットワーク)
WORKER_ENABLED=1 の場合、このノードは EvoMap ネットワーク のワーカーとして参加します。ハートビート経由で機能を公開し、ネットワークの利用可能な作業キューからタスクを取得します。タスクは進化サイクルの成功後の solidify 中にアトミックにクレームされます。
| 変数 | デフォルト | 説明 |
|---|---|---|
WORKER_ENABLED | _(未設定)_ | 1 に設定してワーカープールモードを有効化 |
WORKER_DOMAINS | _(空)_ | このワーカーが受け入れるタスクドメインのカンマ区切りリスト (例: repair,harden) |
WORKER_MAX_LOAD | 5 | ハブ側スケジューリング用に公開される最大同時タスク容量(ローカルで強制される同時実行制限ではない) |
WORKER_ENABLED=1 WORKER_DOMAINS=repair,harden WORKER_MAX_LOAD=3 evolver --loopWORKER_ENABLED と Website のトグル
evomap.ai のダッシュボードにはノード詳細ページに「Worker」トグルがあります。両者の関係は次のとおりです:
| 制御 | スコープ | 動作 |
|---|---|---|
WORKER_ENABLED=1 (環境変数) | ローカル | ローカル evolver デーモンにハートビートでワーカーメタデータを含めてタスクを受け入れるよう指示 |
| Website トグル | Hub 側 | Hub にこのノードへタスクをディスパッチするかどうかを指示 |
ノードがネットワークからタスクを受け取って実行するには両方が有効である必要があります。どちらかがオフの場合、ノードはネットワークから作業を取得しません。推奨フロー:
1. .env に WORKER_ENABLED=1 を設定し、evolver --loop を開始します。 2. evomap.ai に移動し、自分のノードを見つけて Worker トグルをオンにします。
GEP プロトコル (監査可能な進化)
このリポジトリには GEP (Genome Evolution Protocol) に基づくプロトコル制約付きプロンプトモードが含まれています。
- 構造化アセットは
assets/gep/にあります: assets/gep/genes.jsonassets/gep/capsules.jsonassets/gep/events.jsonl- セレクタロジックは抽出されたシグナルを使用して既存の Gene/Capsule を優先し、プロンプトで JSON セレクタ決定を発行します。
- 制約: ドキュメントで許可されるのは DNA 絵文字のみ。他のすべての絵文字は禁止。
設定と分離
Evolver は環境非依存になるよう設計されています。
コア環境変数
| 変数 | 説明 | デフォルト |
|---|---|---|
EVOLVE_STRATEGY | 進化戦略プリセット (balanced / innovate / harden / repair-only) | balanced |
A2A_HUB_URL | EvoMap Hub URL | _(未設定、オフラインモード)_ |
A2A_NODE_ID | ネットワーク上のノードアイデンティティ | _(デバイスフィンガープリントから自動生成)_ |
HEARTBEAT_INTERVAL_MS | Hub ハートビート間隔 | 360000 (6 分) |
MEMORY_DIR | メモリファイルのパス | ./memory |
EVOLVE_REPORT_TOOL | 結果報告用のツール名 | message |
ローカルオーバーライド(注入)
コアコードを変更せずに、ローカル設定(例: レポートに message の代わりに feishu-card を使用)を注入できます。
方法 1: 環境変数 .env ファイルに EVOLVE_REPORT_TOOL を設定:
EVOLVE_REPORT_TOOL=feishu-card方法 2: 動的検出 スクリプトは、互換性のあるローカルスキル(skills/feishu-card など)がワークスペースに存在するかを自動的に検出し、それに応じて動作をアップグレードします。
バリデータ役割(デフォルト ON)
EvoMap Hub に接続すると、各 evolver インスタンスは分散バリデータとしても動作します:hub から割り当てられた検証タスクを定期的に取得し、提案者が宣言した検証コマンドをサンドボックスで実行し、ValidationReport を返送します。コンセンサスに参加したバリデータはクレジットと評判を獲得します。
| 変数 | デフォルト | 説明 |
|---|---|---|
EVOLVER_VALIDATOR_ENABLED | _(未設定 = ON)_ | 0/false/off でオプトアウト、1/true/on で強制 ON。env が hub プッシュフラグおよびコードのデフォルトより優先されます。 |
EVOLVER_VALIDATOR_DAEMON_INTERVAL_MS | 60000 | --loop/--mad-dog モードでのバリデータ常駐ポーリング間隔。 |
EVOLVER_VALIDATOR_MAX_TASKS_PER_CYCLE | 2 | ポーリングごとの最大取得タスク数。 |
EVOLVER_VALIDATOR_FETCH_TIMEOUT_MS | 8000 | 1 回のフェッチのタイムアウト。 |
永続フラグの上書き:env が未設定の場合、ランタイムは ~/.evomap/feature_flags.json を読み込みます。Hub は既存の mailbox 経由で feature_flag_update イベントを送り、アップグレード後のレガシーノードを自動 ON にできます。
永続的にオプトアウト:
EVOLVER_VALIDATOR_ENABLED=0 evolver --loopGitHub Issue 自動報告
evolver が持続的な失敗(失敗ループまたは高い失敗率での繰り返しエラー)を検出すると、サニタイズされた環境情報とログで GitHub issue を上流リポジトリに自動的にファイルできます。すべての機密データ(トークン、ローカルパス、メールなど)は送信前に編集されます。
| 変数 | デフォルト | 説明 |
|---|---|---|
EVOLVER_AUTO_ISSUE | true | 自動 issue 報告の有効/無効 |
EVOLVER_ISSUE_REPO | EvoMap/evolver | ターゲット GitHub リポジトリ (owner/repo) |
EVOLVER_ISSUE_COOLDOWN_MS | 86400000 (24h) | 同じエラーシグネチャのクールダウン期間 |
EVOLVER_ISSUE_MIN_STREAK | 5 | トリガーする最小連続失敗ストリーク |
repo スコープを持つ GITHUB_TOKEN(または GH_TOKEN / GITHUB_PAT)が必要です。トークンが利用できない場合、機能は静かにスキップされます。
セキュリティモデル
このセクションでは、Evolver の実行境界と信頼モデルについて説明します。
何が実行され、何が実行されないか
| コンポーネント | 動作 | シェルコマンドを実行? |
|---|---|---|
src/evolve.js | ログ読み取り、Gene 選択、プロンプト構築、アーティファクト書き込み | 読み取り専用の git/プロセスクエリのみ |
src/gep/prompt.js | GEP プロトコルプロンプト文字列を組み立て | いいえ(純粋なテキスト生成) |
src/gep/selector.js | シグナルマッチングで Gene/Capsule をスコアリングおよび選択 | いいえ(純粋なロジック) |
src/gep/solidify.js | Gene の validation コマンド経由でパッチを検証 | はい(下記参照) |
index.js (ループ復旧) | クラッシュ時に sessions_spawn(...) テキストを stdout に出力 | いいえ(テキスト出力のみ;実行はホストランタイムに依存) |
Gene 検証コマンドの安全性
solidify.js は Gene の validation 配列に列挙されたコマンドを実行します。任意のコマンド実行を防ぐため、すべての検証コマンドは安全性チェック (isValidationCommandAllowed) によってゲートされています:
1. プレフィックスホワイトリスト: node、npm、npx で始まるコマンドのみ許可。 2. コマンド置換なし: バッククォートと $(...) はコマンド文字列のどこでも拒否。 3. シェル演算子なし: 引用されたコンテンツを削除した後、;、&、|、>、< は拒否。 4. タイムアウト: 各コマンドは 180 秒に制限。 5. スコープ実行: コマンドは cwd をリポジトリルートに設定して実行。
A2A 外部アセット取り込み
scripts/a2a_ingest.js 経由で取り込まれた外部 Gene/Capsule アセットは、分離された候補ゾーンにステージングされます。ローカルストア (scripts/a2a_promote.js) への昇格には次が必要です:
1. 明示的な --validated フラグ(オペレータが最初にアセットを検証する必要がある)。 2. Gene の場合: すべての validation コマンドは昇格前に同じ安全性チェックに対して監査されます。安全でないコマンドは昇格を拒否されます。 3. Gene 昇格は、同じ ID の既存のローカル Gene を決して上書きしません。
sessions_spawn 出力
index.js と evolve.js の sessions_spawn(...) 文字列は、直接の関数呼び出しではなく、stdout へのテキスト出力です。これらが解釈されるかどうかはホストランタイム(例: OpenClaw プラットフォーム)に依存します。evolver 自体は sessions_spawn を実行可能コードとして呼び出しません。
バージョニング (SemVer)
MAJOR.MINOR.PATCH
- MAJOR: 互換性のない変更
- MINOR: 後方互換性のある機能
- PATCH: 後方互換性のあるバグ修正
変更履歴
完全なリリース履歴は GitHub Releases をご覧ください。
FAQ
これはコードを自動的に編集しますか? いいえ。Evolver は進化をガイドするプロトコル束縛のプロンプトとアセットを生成します。ソースコードを直接変更することはありません。Evolver ができること・できないこと を参照してください。
`evolver --loop` を実行したが、テキストを出力し続けるだけです。動作していますか? はい。スタンドアロンモードでは、evolver は GEP プロンプトを生成して stdout に出力します。変更を自動的に適用すると期待した場合は、出力を解釈する OpenClaw のようなホストランタイムが必要です。または、--review モードを使用して各進化ステップを手動でレビューして適用します。
EvoMap Hub への接続は必要ですか? いいえ。すべてのコア進化機能はオフラインで動作します。Hub 接続は、スキルストア、ワーカープール、進化リーダーボードなどのネットワーク機能にのみ必要です。EvoMap Hub への接続 を参照してください。
すべての GEP アセットを使用する必要がありますか? いいえ。デフォルトの Gene から始めて、時間をかけて拡張できます。
本番環境で安全ですか? レビューモードと検証ステップを使用してください。ライブパッチャーではなく、安全性重視の進化ツールとして扱ってください。セキュリティモデル を参照してください。
このリポジトリはどこにクローンすべきですか? 任意のディレクトリにクローンします。OpenClaw を使用する場合は、ホストランタイムが evolver の stdout にアクセスできるよう、OpenClaw ワークスペースにクローンします。スタンドアロン使用の場合、任意の場所で動作します。
ロードマップ
方針であり確約ではありません。最新のバックログは GitHub Issues を参照してください。
- オンボーディング: 1 分間のクイックスタートデモと、他のエージェント進化手法との比較表。
- GEP 統合の深化: より豊富なシグナル抽出と Gene / Capsule 選択、および再利用アナリティクス。
- メモリとスキル: セッション結果を再利用可能な Gene / Capsule へより速く蒸留。
- 対応ランタイムの拡大: Cursor / Claude Code / Codex / Kiro / opencode / OpenClaw 以外のホスト統合を拡充。
Star 履歴

謝辞
- onthebigtree -- evomap 進化ネットワークの作成にインスピレーションを与えた。3 つのランタイムおよびロジックバグを修正(PR #25)。ホスト名プライバシーハッシュ、ポータブルな検証パス、デッドコードクリーンアップに貢献(PR #26)。
- lichunr -- 私たちのコンピュートネットワークが無料で使用するために数千ドル相当のトークンを提供。
- shinjiyu -- 多数のバグレポートを提出し、スニペット付きタグを持つ多言語シグナル抽出に貢献(PR #112)。
- voidborne-d -- 11 の新しい認証情報編集パターンでブロードキャスト前のサニタイズを強化(PR #107)。strategy、validationReport、envFingerprint のために 45 のテストを追加(PR #139)。
- blackdogcat -- 欠落していた dotenv 依存関係を修正し、インテリジェントな CPU 負荷閾値自動計算を実装(PR #144)。
- LKCY33 -- .env 読み込みパスとディレクトリ権限を修正(PR #21)。
- hendrixAIDev -- ドライランモードで performMaintenance() が実行される問題を修正(PR #68)。
- toller892 -- events.jsonl forbidden_paths バグを独立に特定して報告(PR #149)。
- WeZZard -- SKILL.md に A2A_NODE_ID セットアップガイドを追加し、NODE_ID が明示的に設定されていない場合に a2aProtocol でコンソール警告を追加(PR #164)。
- Golden-Koi -- README に cron/外部ランナーキープアライブのベストプラクティスを追加(PR #167)。
- upbit -- evolver および evomap 技術の普及に重要な役割を果たした。
- Chi Jianqiang -- プロモーションとユーザー体験の改善に多大な貢献。
ライセンス
コア進化エンジンモジュールは、知的財産を保護するために難読化された形式で配布されます。ソース: EvoMap/evolver。
const { loadGenes, loadCapsules, readAllEvents } = require('../src/gep/assetStore');
const { exportEligibleCapsules, exportEligibleGenes, isAllowedA2AAsset } = require('../src/gep/a2a');
const { buildPublish, buildHello, getTransport } = require('../src/gep/a2aProtocol');
const { computeAssetId, SCHEMA_VERSION } = require('../src/gep/contentHash');
function main() {
var args = process.argv.slice(2);
var asJson = args.includes('--json');
var asProtocol = args.includes('--protocol');
var withHello = args.includes('--hello');
var persist = args.includes('--persist');
var includeEvents = args.includes('--include-events');
var capsules = loadCapsules();
var genes = loadGenes();
var events = readAllEvents();
// Build eligible list: Capsules (filtered) + Genes (filtered) + Events (opt-in)
var eligibleCapsules = exportEligibleCapsules({ capsules: capsules, events: events });
var eligibleGenes = exportEligibleGenes({ genes: genes });
var eligible = eligibleCapsules.concat(eligibleGenes);
if (includeEvents) {
var eligibleEvents = (Array.isArray(events) ? events : []).filter(function (e) {
return isAllowedA2AAsset(e) && e.type === 'EvolutionEvent';
});
for (var ei = 0; ei < eligibleEvents.length; ei++) {
var ev = eligibleEvents[ei];
if (!ev.schema_version) ev.schema_version = SCHEMA_VERSION;
if (!ev.asset_id) { try { ev.asset_id = computeAssetId(ev); } catch (e) {} }
}
eligible = eligible.concat(eligibleEvents);
}
if (withHello || asProtocol) {
var hello = buildHello({ geneCount: genes.length, capsuleCount: capsules.length });
process.stdout.write(JSON.stringify(hello) + '\n');
if (persist) { try { getTransport().send(hello); } catch (e) {} }
}
if (asProtocol) {
for (var i = 0; i < eligible.length; i++) {
var msg = buildPublish({ asset: eligible[i] });
process.stdout.write(JSON.stringify(msg) + '\n');
if (persist) { try { getTransport().send(msg); } catch (e) {} }
}
return;
}
if (asJson) {
process.stdout.write(JSON.stringify(eligible, null, 2) + '\n');
return;
}
for (var j = 0; j < eligible.length; j++) {
process.stdout.write(JSON.stringify(eligible[j]) + '\n');
}
}
try { main(); } catch (e) {
process.stderr.write((e && e.message ? e.message : String(e)) + '\n');
process.exit(1);
}
var fs = require('fs');
var assetStore = require('../src/gep/assetStore');
var a2a = require('../src/gep/a2a');
var memGraph = require('../src/gep/memoryGraphAdapter');
var contentHash = require('../src/gep/contentHash');
var a2aProto = require('../src/gep/a2aProtocol');
function readStdin() {
try { return fs.readFileSync(0, 'utf8'); } catch (e) { return ''; }
}
function parseSignalsFromEnv() {
var raw = process.env.A2A_SIGNALS || '';
if (!raw) return [];
try {
var maybe = JSON.parse(raw);
if (Array.isArray(maybe)) return maybe.map(String).filter(Boolean);
} catch (e) {}
return String(raw).split(',').map(function (s) { return s.trim(); }).filter(Boolean);
}
function main() {
var args = process.argv.slice(2);
var inputPath = '';
for (var i = 0; i < args.length; i++) {
if (args[i] && !args[i].startsWith('--')) { inputPath = args[i]; break; }
}
var source = process.env.A2A_SOURCE || 'external';
var factor = Number.isFinite(Number(process.env.A2A_EXTERNAL_CONFIDENCE_FACTOR))
? Number(process.env.A2A_EXTERNAL_CONFIDENCE_FACTOR) : 0.6;
var text = inputPath ? a2a.readTextIfExists(inputPath) : readStdin();
var parsed = a2a.parseA2AInput(text);
var signals = parseSignalsFromEnv();
var accepted = 0;
var rejected = 0;
var emitDecisions = process.env.A2A_EMIT_DECISIONS === 'true';
for (var j = 0; j < parsed.length; j++) {
var obj = parsed[j];
if (!a2a.isAllowedA2AAsset(obj)) continue;
if (obj.asset_id && typeof obj.asset_id === 'string') {
if (!contentHash.verifyAssetId(obj)) {
rejected += 1;
if (emitDecisions) {
try {
var dm = a2aProto.buildDecision({ assetId: obj.asset_id, localId: obj.id, decision: 'reject', reason: 'asset_id integrity check failed' });
a2aProto.getTransport().send(dm);
} catch (e) {}
}
continue;
}
}
var staged = a2a.lowerConfidence(obj, { source: source, factor: factor });
if (!staged) continue;
assetStore.appendExternalCandidateJsonl(staged);
try { memGraph.recordExternalCandidate({ asset: staged, source: source, signals: signals }); } catch (e) {}
if (emitDecisions) {
try {
var dm2 = a2aProto.buildDecision({ assetId: staged.asset_id, localId: staged.id, decision: 'quarantine', reason: 'staged as external candidate' });
a2aProto.getTransport().send(dm2);
} catch (e) {}
}
accepted += 1;
}
process.stdout.write('accepted=' + accepted + ' rejected=' + rejected + '\n');
}
try { main(); } catch (e) {
process.stderr.write((e && e.message ? e.message : String(e)) + '\n');
process.exit(1);
}
var assetStore = require('../src/gep/assetStore');
var solidifyMod = require('../src/gep/solidify');
var contentHash = require('../src/gep/contentHash');
var a2aProto = require('../src/gep/a2aProtocol');
function parseArgs(argv) {
var out = { flags: new Set(), kv: new Map(), positionals: [] };
for (var i = 0; i < argv.length; i++) {
var a = argv[i];
if (!a) continue;
if (a.startsWith('--')) {
var eq = a.indexOf('=');
if (eq > -1) { out.kv.set(a.slice(2, eq), a.slice(eq + 1)); }
else {
var key = a.slice(2);
var next = argv[i + 1];
if (next && !String(next).startsWith('--')) { out.kv.set(key, next); i++; }
else { out.flags.add(key); }
}
} else { out.positionals.push(a); }
}
return out;
}
function main() {
var args = parseArgs(process.argv.slice(2));
var id = String(args.kv.get('id') || '').trim();
var typeRaw = String(args.kv.get('type') || '').trim().toLowerCase();
var validated = args.flags.has('validated') || String(args.kv.get('validated') || '') === 'true';
var limit = Number.isFinite(Number(args.kv.get('limit'))) ? Number(args.kv.get('limit')) : 500;
if (!id || !typeRaw) throw new Error('Usage: node scripts/a2a_promote.js --type capsule|gene|event --id <id> --validated');
if (!validated) throw new Error('Refusing to promote without --validated (local verification must be done first).');
var type = typeRaw === 'capsule' ? 'Capsule' : typeRaw === 'gene' ? 'Gene' : typeRaw === 'event' ? 'EvolutionEvent' : '';
if (!type) throw new Error('Invalid --type. Use capsule, gene, or event.');
var external = assetStore.readRecentExternalCandidates(limit);
var candidate = null;
for (var i = 0; i < external.length; i++) {
if (external[i] && external[i].type === type && String(external[i].id) === id) { candidate = external[i]; break; }
}
if (!candidate) throw new Error('Candidate not found in external zone: type=' + type + ' id=' + id);
if (type === 'Gene') {
var validation = Array.isArray(candidate.validation) ? candidate.validation : [];
for (var j = 0; j < validation.length; j++) {
var c = String(validation[j] || '').trim();
if (!c) continue;
if (!solidifyMod.isValidationCommandAllowed(c)) {
throw new Error('Refusing to promote Gene ' + id + ': validation command rejected by safety check: "' + c + '". Only node/npm/npx commands without shell operators are allowed.');
}
}
}
var promoted = JSON.parse(JSON.stringify(candidate));
if (!promoted.a2a || typeof promoted.a2a !== 'object') promoted.a2a = {};
promoted.a2a.status = 'promoted';
promoted.a2a.promoted_at = new Date().toISOString();
if (!promoted.schema_version) promoted.schema_version = contentHash.SCHEMA_VERSION;
promoted.asset_id = contentHash.computeAssetId(promoted);
var emitDecisions = process.env.A2A_EMIT_DECISIONS === 'true';
if (type === 'EvolutionEvent') {
assetStore.appendEventJsonl(promoted);
if (emitDecisions) {
try {
var dmEv = a2aProto.buildDecision({ assetId: promoted.asset_id, localId: id, decision: 'accept', reason: 'event promoted for provenance tracking' });
a2aProto.getTransport().send(dmEv);
} catch (e) {}
}
process.stdout.write('promoted_event=' + id + '\n');
return;
}
if (type === 'Capsule') {
assetStore.appendCapsule(promoted);
if (emitDecisions) {
try {
var dm = a2aProto.buildDecision({ assetId: promoted.asset_id, localId: id, decision: 'accept', reason: 'capsule promoted after validation' });
a2aProto.getTransport().send(dm);
} catch (e) {}
}
process.stdout.write('promoted_capsule=' + id + '\n');
return;
}
var localGenes = assetStore.loadGenes();
var exists = false;
for (var k = 0; k < localGenes.length; k++) {
if (localGenes[k] && localGenes[k].type === 'Gene' && String(localGenes[k].id) === id) { exists = true; break; }
}
if (exists) {
if (emitDecisions) {
try {
var dm2 = a2aProto.buildDecision({ assetId: promoted.asset_id, localId: id, decision: 'reject', reason: 'local gene with same ID already exists' });
a2aProto.getTransport().send(dm2);
} catch (e) {}
}
process.stdout.write('conflict_keep_local_gene=' + id + '\n');
return;
}
assetStore.upsertGene(promoted);
if (emitDecisions) {
try {
var dm3 = a2aProto.buildDecision({ assetId: promoted.asset_id, localId: id, decision: 'accept', reason: 'gene promoted after safety audit' });
a2aProto.getTransport().send(dm3);
} catch (e) {}
}
process.stdout.write('promoted_gene=' + id + '\n');
}
try { main(); } catch (e) {
process.stderr.write((e && e.message ? e.message : String(e)) + '\n');
process.exit(1);
}
// Canary script: run in a forked child process to verify index.js loads
// without crashing. Exit 0 = safe, non-zero = broken.
//
// This is the last safety net before solidify commits an evolution.
// If a patch broke index.js (syntax error, missing require, etc.),
// the canary catches it BEFORE the daemon restarts with broken code.
try {
require('../index.js');
process.exit(0);
} catch (e) {
process.stderr.write(String(e.message || e).slice(0, 500));
process.exit(1);
}
Related skills
Forks & variants (2)
Capability Evolver has 2 known copies in the catalog totaling 4k installs. They canonicalize to this original listing.
- autogame-17 - 4k installs
- bighardperson - 8 installs
How it compares
capability-evolver implements its own SKILL.md workflow rather than a generic substitute skill.
FAQ
Who is capability-evolver for?
Agents and developers following the capability-evolver SKILL.md guidance.
When should I use capability-evolver?
When user intent matches description triggers and quick start scenarios.
Is capability-evolver safe to install?
Review the Security Audits panel before production shell or network use.