charon-fan/agent-playbook
24 skills49k installs1.6k starsGitHub
Install
npx skills add https://github.com/charon-fan/agent-playbookSkills in this repo
1Self Improving AgentSelf-Improving Agent is a universal self-improvement system for AI agents that learns from all skill experiences. Agents use it to accumulate patterns, detect and fix guidance errors, and continuously evolve their own capabilities. It matters because agents can validate and improve their own instructions over time - detected patterns become semantic memory, failed assumptions get corrected with evidence markers, and reusable knowledge promotes into skill updates.32.4kinstalls2Planning With FilesThe planning-with-files skill uses persistent markdown files for general planning, progress tracking, and knowledge storage (Manus-style workflow). Use for multi-step tasks, research projects, or general organization WITHOUT mentioning PRD. For PRD-specific work, use prd-planner skill instead. # Planning with Files > "Work like Manus" - Uses persistent markdown files for planning, progress tracking, and knowledge storage. ## Description A Claude Code skill that transforms your workflow to use persistent markdown files for planning and progress tracking - the pattern that made Manus AI worth billions. ## The Problem Claude Code (and most AI agents) suffer from: - **Volatile memory** - TodoWrite tool disappears on context reset - **Goal drift** - Original goals get forgotten after many tool calls - **Hidden errors** - Failures aren't tracked, mistakes repeat - **Context stuffing** - Everything crammed into context instead of stored ## The Solution: 3-File Pattern For every complex task, create THREE files: ```text task_plan.md → Track phases and progress notes.md → Store research and findings [deliverable].md → Final output ``` ### The Workflow Loop ```text 1.1.1kinstalls3Security Auditorsecurity-auditor is a security vulnerability expert skill from charon-fan/agent-playbook focused on OWASP Top 10 coverage and common application security issues. It activates when developers request security audits, mention security review, or review code for vulnerabilities before deployment. The skill uses Read, Grep, Glob, Bash, and WebSearch tools to inspect codebases and logs audit patterns via session hooks. Developers reach for security-auditor as a structured pre-ship gate when they need OWASP-aligned findings rather than ad-hoc grep for obvious mistakes. It pairs naturally with code review workflows where exploitable flaws must be caught before merge or production rollout.981installs4Architecting Solutionsarchitecting-solutions is a charon-fan/agent-playbook skill for technical solution and architecture design in Claude Code. It clarifies requirements and success criteria, analyzes constraints and existing patterns, proposes options with trade-offs, and documents the decision in a PRD-style output under docs/. Example triggers include designing a billing system, multi-tenant analytics architecture, or a data migration plan. Developers reach for this skill at the start of substantial features when requirements are still fuzzy and multiple architectural paths exist. The output gives a decision record teams can implement against instead of debating options ad hoc in pull requests.814installs5Skill Routerskill-router is an intelligent routing layer from charon-fan/agent-playbook for Claude Code skills. It pipelines user requests through intent analysis, semantic skill matching, and optional interactive clarification before recommending the most appropriate skill. Installation symlinks the skill into ~/.claude/skills/skill-router. Developers reach for skill-router when they describe a task in natural language—fix my tests, deploy to Vercel, write docs—and want the agent to pick the right playbook instead of guessing or loading the wrong skill. It is catalog-aware infrastructure, not a domain implementation skill itself.801installs6Prd Plannerprd-planner is a PRD creation skill in charon-fan/agent-playbook that combines PRD methodology with file-based persistence so agents do not lose context while drafting requirements. Traditional PRD workflows suffer from forgotten research, mode switching, and inconsistent documents; prd-planner addresses this with a 4-file pattern that stores thinking on disk across research, synthesis, and technical decision steps. Developers reach for prd-planner when starting a new feature, aligning stakeholders, or handing a coherent spec to coding agents. The workflow produces a single coherent PRD instead of fragmented chat memory. prd-planner is suited to agent-assisted product work where multiple sessions contribute to the same requirements doc.772installs7Test Automatortest-automator is a test generation skill from charon-fan/agent-playbook that creates and maintains automated unit and integration tests with framework-specific patterns. The skill supports 7 test frameworks: Jest, Vitest, and Mocha for TypeScript/JavaScript, pytest and unittest for Python, Go's testing package, and JUnit for Java. Developers reach for test-automator when writing tests for new functions, improving coverage on existing modules, or scaffolding test boilerplate. The skill bundles 2 Python scripts—generate_test.py for boilerplate generation and coverage_report.py for coverage analysis—invoked from the command line alongside conversational test authoring.758installs8Figma Designerfigma-designer is a skill in charon-fan/agent-playbook that connects to the Figma MCP server to analyze frames and emit detailed product requirement documents with pixel-perfect visual specifications. It extracts design tokens, component specifications, and layout measurements developers can implement without re-measuring in the Figma UI. Installation uses a symbolic link from agent-playbook/skills/figma-designer/SKILL.md into the Claude Code skills directory. Prerequisites include a running Figma MCP server the agent can query. Reach for figma-designer when a Figma file is the source of truth and you need structured PRD output with spacing, typography tokens, and component boundaries before writing React, Swift, or other frontend code.754installs9Workflow OrchestratorWorkflow Orchestrator is a Claude Code skill from charon-fan/agent-playbook that coordinates multi-skill workflows using hook definitions. It reads hook definitions from skills/auto-trigger/SKILL.md and executes follow-up actions based on auto, background, or ask_first modes while logging progress and context to session-logger. Developers invoke it with phrases like Complete workflow or Finish the process and trigger next steps after milestones such as PRD creation or implementation. Workflow Orchestrator removes manual skill handoffs when an agent playbook defines sequential or conditional next actions. It fits teams building repeatable agent pipelines where one completed task should reliably launch the next skill without re-prompting.753installs10Qa Expertqa-expert is a quality assurance specialist skill for coding agents that helps developers design risk-based testing strategies, balance the testing pyramid, and define enforceable quality gates before writing tests. The skill activates when teams ask for QA strategy, coverage planning, or quality standards and can use Read, Write, Edit, Bash, Grep, Glob, WebFetch, and WebSearch to research practices and draft gates. A session-logger hook records QA activity after completion. Developers reach for qa-expert when they need a testing plan—not individual test cases—covering what to test, at which layer, and what must pass before merge or release.702installs11Session Loggersession-logger is a Claude Code skill from charon-fan/agent-playbook that automatically saves agent conversations to persistent markdown files under a sessions/ directory. Installation symlinks the skill into ~/.claude/skills/session-logger. Trigger phrases in Chinese and English—such as 保存对话信息—cause the agent to review the conversation, extract key decisions, and write a structured session log without manual copy-paste. Developers reach for session-logger when long agent threads contain architecture choices, debugging paths, or TODOs that would be costly to reconstruct in the next session. Logs support referencing prior decisions, tracking project evolution, and learning from earlier problem-solving approaches across Claude Code runs.699installs12Debuggerdebugger is an advanced debugging specialist skill for coding agents that walks developers through a systematic reproduce-isolate-root-cause-fix-verify loop when bugs or unexpected behavior appear. The skill activates on error reports, stack traces, failing tests, or explicit debugging requests and uses Read, Write, Edit, Bash, Grep, and Glob to inspect code and validate fixes. After completion, hooks can trigger self-improving-agent learning and session logging to capture debugging patterns. Developers reach for debugger when ad-hoc guessing stalls and they need a repeatable diagnostic workflow inside their agent session.698installs13Code Reviewercode-reviewer is an agent-playbook skill for comprehensive review of pull requests and local diffs in Claude Code. The skill analyzes changes across correctness, security including OWASP Top 10 and injection prevention, performance, and test coverage, then returns structured feedback developers can act on before merge. Invoke it with prompts like Review this PR, Check my changes, or Review the code in src/auth/. The workflow follows five review steps from change analysis through categorized findings, making it a repeatable pre-merge gate for teams without a dedicated reviewer available.690installs14Refactoring Specialistrefactoring-specialist is a Claude Code skill in charon-fan/agent-playbook for behavior-preserving code cleanup and technical debt reduction. It maps common code smells—Long Method, Duplicate Code, Large Class, Long Parameter List, and Switch Statement—to named refactorings such as Extract Method, Extract Class, Introduce Parameter Object, and Replace with Polymorphism. Developers invoke it with prompts like refactor this code, clean up this function, or this code is messy when they want guided small-step improvements instead of risky rewrites. The skill enforces behavior preservation and incremental changes as core principles for safe refactors across languages.686installs15Prd Implementation Precheckprd-implementation-precheck is an agent-playbook skill that performs mandatory preflight review before implementing any PRD. Instead of blindly coding from a product requirements document, the skill reviews scope, alignment, dependencies, and risks, presents findings and questions to the user, waits for confirmation, then implements with minimal consistent changes. Developers reach for prd-implementation-precheck when asked to implement a PRD file—such as docs/feature-prd.md—and want issues caught before coding begins. Installation symlinks the skill into ~/.claude/skills for Claude Code sessions.668installs16Api Designerapi-designer is an agent-playbook skill that acts as a REST and GraphQL API architect for new designs or reviews of existing APIs. It emphasizes resource-oriented routes, correct HTTP verbs, meaningful status codes, and maintainable specifications developers can implement against. The skill allows Read, Write, Edit, Bash, Grep, Glob, WebFetch, and WebSearch, and triggers background hooks for pattern learning and session logging after completion. Reach for api-designer when scaffolding a new service boundary, reviewing an API for scalability, or improving an existing surface before implementation begins.659installs17Api Documenterapi-documenter is an agent-playbook skill for producing OpenAPI 3.0 and Swagger-compatible API documentation from existing REST services. It follows RESTful conventions with clear resource naming, complete request and response schemas, authentication requirements, and standardized error formats. Bundled Python scripts generate specs via generate_openapi.py and validate them with validate_openapi.py against openapi.yaml. Developers invoke api-documenter when they need machine-readable API contracts for SDK generation, partner integrations, or developer portals.659installs18Performance Engineerperformance-engineer is a Claude Code skill from charon-fan/agent-playbook for performance optimization and analysis when developers ask to optimize code, explain slowness, or profile an application. It defines explicit targets: API response p50 under 100ms, p95 under 500ms, database queries under 50ms, First Meaningful Paint under 2 seconds, and Time to Interactive under 3 seconds. Bundled Python scripts include scripts/profile.py for application profiling and scripts/perf_report.py for generating performance reports. Developers reach for performance-engineer when backend services, APIs, or frontend experiences miss latency budgets and need structured diagnosis rather than ad-hoc micro-optimizations.654installs19Documentation Engineerdocumentation-engineer is an agent-playbook skill for creating clear, comprehensive documentation across README overviews, API endpoint references, inline code comments, and architecture write-ups. The collection includes Python helpers such as scripts/generate_docs.py for scaffolding structures and validation scripts for doc quality checks. Developers invoke it with prompts like write documentation for this API, create a README, or document this code. Reach for documentation-engineer when onboarding docs lag behind implementation or when API surfaces need structured references before release.652installs20Auto Triggerauto-trigger is a configuration skill in the charon-fan/agent-playbook collection that declares automatic trigger relationships between skills via YAML front-matter hooks. It is not invoked directly; workflow-orchestrator reads these hook definitions and executes follow-up skills such as session-logger when a parent skill completes. Developers reach for auto-trigger when building repeatable agent workflows where one skill should always chain into logging, validation, or cleanup steps. Example hooks use after_complete with mode auto to fire downstream skills without extra user input.644installs21Commit Helpercommit-helper is a Claude Code skill from charon-fan/agent-playbook in the agent-playbook collection. When asked to commit changes, the skill runs `git diff`, analyzes the patch, drafts a Conventional Commits message with type, optional scope, and subject line, presents it for approval, and executes the commit on confirmation. Examples include scoped messages like `feat(auth): add OAuth2 login support` for feature work. Developers reach for commit-helper during daily Git workflows when they want consistent commit semantics across teams following the Conventional Commits specification without writing messages manually.636installs22Deployment EngineerDeployment Engineer is a Claude Code skill in charon-fan/agent-playbook for CI/CD pipeline design and deployment automation. It helps developers scaffold GitHub Actions workflows, Kubernetes manifests, and environment-specific deploy configs while choosing among blue-green, rolling, and canary strategies. The skill ships Python helpers—generate_deploy.py for environment configs and validate_deploy.py for pre-flight checks—so agents produce reviewable artifacts instead of hand-waving YAML. Developers reach for Deployment Engineer when standing up first pipelines, standardizing multi-environment releases, or encoding zero-downtime rollback patterns. Trigger phrases include “set up CI/CD,” “create deployment pipeline,” and “configure GitHub Actions.” Outputs are validated pipeline files, strategy rationale, and deployment configs ready for repository commit.633installs23Create Prcreate-pr streamlines pull request creation for the charon-fan/agent-playbook repository with automatic change analysis and bilingual documentation sync. The skill examines git diff to understand code changes, updates both README.md and README.zh-CN.md to maintain English–Chinese parity, fills a consistent PR description template, and runs a verification checklist before submission. Developers reach for create-pr when submitting agent-playbook changes and need EN/ZH docs kept synchronized without manual diff review and README editing on every PR.625installs24Long Task Coordinatorlong-task-coordinator is a skill from charon-fan/agent-playbook for agent workflows that outlive a single chat window. It persists progress in a durable state file, recovers status before proposing new steps, splits work into worker-sized chunks, and reports honestly when execution is awaiting external results instead of falsely claiming completion. Acceptance criteria prompts in the eval suite cover resuming interrupted migrations and delegating research with visible ownership tracking. Developers reach for long-task-coordinator when coding agents might lose context across sessions or prematurely mark large migrations and research efforts as done.624installs