rohitg00/agentmemory
14 skills8.4k installs307k starsGitHub
Install
npx skills add https://github.com/rohitg00/agentmemorySkills in this repo
1Rememberremember is a thin agent skill for the agentmemory plugin that turns explicit user requests into durable long-term memories via the memory_save MCP tool. Solo builders running Claude Code with agentmemory enabled use it whenever a session surfaces a decision, workaround, or architecture fact that should survive the next chat—without maintaining a separate notes app. The skill instructs the agent to distill the user’s wording, extract precise concept keywords (for example jwt-refresh-rotation instead of auth), attach relevant file paths, and confirm what was stored so retrieval later is predictable. It is beginner-friendly procedural glue: no custom server code, but it assumes the plugin’s .mcp.json loaded on startup. Journey-wide because insights arise during Idea research, Build implementation, Ship incidents, and Operate maintenance alike. It does not replace structured docs or git history; it complements episodic agent context with searchable memory entries.1.3kinstalls2RecallThe recall skill teaches an agent how to pull structured memories out of AgentMemory using memory_smart_search when a solo builder asks what was decided earlier, how code was shaped, or what was deferred. It is journey-wide procedural knowledge: the same invocation pattern applies while you are still in Build debating caching, during Ship when reviewing a limiter implementation, or during Operate when diagnosing regressions against past choices. The readme walks through JSON request and response shapes, including importance weighting and memory types, and shows how to phrase answers in plain language that cites session identifiers. It does not write new memories or mutate sessions—it is a read path optimized for Q&A. Builders running long-lived Claude Code or Cursor projects benefit when architectural decisions would otherwise live only in scrolled-away chat. Confidence is high on the search-and-present workflow; linking to the truncated fallback command may require reading the full repo SKILL.md.1.2kinstalls3Session Historysession-history connects solo builders to the agentmemory plugin’s MCP `memory_sessions` tool so you can see what actually happened in past Claude Code sessions on the current project. Use it when you return after a day away, hand off mentally to your agent, or need a clean timeline of observations and decisions instead of scrolling chat fragments. The skill specifies fetch parameters, presentation order, and how to surface summaries and observation counts without inventing entries—a common failure mode for agents asked about history. It assumes the agentmemory server is wired through the plugin’s `.mcp.json` on startup. When the server is down, it gives concrete recovery steps rather than hallucinating prior work.1.2kinstalls4Commit Contextcommit-context is an agentmemory skill that connects git history to recorded coding sessions. Solo builders use it when a teammate—or a past agent run—left code without a story: you run git blame or git log for the symbol or path, pass the SHA into memory_commit_lookup, and read which session and high-importance observations justified the change. When linking is missing, the skill still orients you with commit message and diff context instead of inventing rationale. It fits repos that already log agent work through agentmemory, especially auth, middleware, and long-lived services where retries and guards accumulate. Expect shell access to git and whichever API backs memory_commit_lookup; outputs are human-readable narratives you can paste into PRs, runbooks, or follow-up tasks.1.2kinstalls5Recaprecap is a user-invocable AgentMemory skill that compresses scattered agent sessions into a dated digest for the repo you are in now. After parsing time windows from arguments, it calls memory_sessions, matches cwd to the working project, sorts by startedAt, and groups sessions under YYYY-MM-DD headings with short ids, titles, observation counts, and status. For each session it optionally recalls up to three high-importance observations so you see decisions and blockers—not just session metadata. Solo builders use it when returning from a weekend, onboarding a fresh agent thread, or answering “what have we been doing this week?” without manually scrolling chat logs. It does not fabricate history; empty windows get an explicit empty message.1.2kinstalls6Commit Historycommit-history is a user-invocable AgentMemory skill that answers “what did my agent commit?” by querying linked git history. Solo builders running long agent sessions need a trustworthy ledger: short SHAs, branches, first-line messages, linked session identifiers, and observation counts—without hallucinated entries. The skill parses optional branch, repo fragment, and limit tokens from arguments, calls the memory_commits MCP tool when available, and falls back to authenticated HTTP against AGENTMEMORY_URL with properly encoded query parameters so special characters in repo URLs do not break requests. Empty filters get actionable guidance to widen the search. It complements generic git log by anchoring commits to agent session context for retrospectives, support, and shipping accountability.1.2kinstalls7Handoffhandoff is a journey-wide agent skill from the agentmemory project that reconnects a cold chat to the most relevant prior agent session for your project directory. Solo builders who context-switch between Idea research, Validate spikes, and long Build sprints install it so Claude Code, Cursor, or similar agents do not re-derive decisions from scratch when they hear resume or pick up where I left off. The procedure calls the memory_sessions MCP tool, ranks sessions by recency and cwd compatibility using strict directory-boundary rules to avoid false matches across similarly named paths, and prioritizes completed work over abandoned threads. If the last turn left a user-facing question unanswered, that question leads the recap before file lists and error history. It is not a replacement for git history or issue trackers; it is short-term procedural memory for agent conversations. You need the memory_sessions MCP available and a recorded session for the repo; without that stack the skill has nothing to load.1.2kinstalls8Agentmemory Agentsagentmemory-agents documents how to attach the AgentMemory server to the coding agent you already use as a solo or indie builder. The connect CLI subcommand targets seventeen named hosts—including Claude Code, Cursor, Codex CLI, Cline, Continue, GitHub Copilot CLI, Gemini CLI, Antigravity, and Droid—with protocol-specific instructions such as MCP-only Cursor, optional hooks on Claude Code and Codex, and Factory Droid ~/.factory/mcp.json paths. Instead of copying stale MCP snippets from a blog post, you follow the generated adapter table aligned with scripts/skills/generate.ts so new agents land in the repo and the skill together. Use during initial project setup in Build, and again in Operate when you migrate laptops, add a second IDE, or enable Codex hooks on desktop. Memory bridge continues to serve REST on :3111 underneath MCP; pair with agentmemory-rest-api when you debug HTTP directly.3installs9Agentmemory Architectureagentmemory-architecture is an agent skill that documents how the agentmemory memory server is built for coding agents: it runs locally on the iii engine, registers mem::* functions and api::* HTTP triggers, and never bypasses that routing model when you add capability. Solo builders install it when they need a precise mental model before changing storage, retrieval, or ports—not when they only want a quick “remember this” command without touching internals. The skill spells out hybrid retrieval combining BM25, vectors, and graph-linked concepts, on-device embeddings for keyless defaults, and optional LLM-backed summaries and auto-injection. It also covers memory fields (content, concepts, files, importance, timestamps), sessions, commit linkage, and lifecycle stages from capture through forget. Use it during Build while wiring agent tooling, and again in Operate when traces or recall quality look wrong. It is reference architecture prose (user-invocable false), ideal to attach before refactors or when answering “how does recall work under the hood?” for your agent.3installs10Agentmemory Configagentmemory-config is a reference skill for solo builders running AgentMemory who need accurate environment tuning without spelunking the TypeScript codebase. It surfaces an autogenerated block of thirty-four AGENTMEMORY_* variables discovered by scanning src/, explains that configuration loads from the environment and from ~/.agentmemory/.env, and flags maintenance via npm run skills:gen when variables change. Topics include agent scope, MCP and Copilot blocks, Docker mode, LLM timeouts, image embedding stores, reflection, and viewer host settings—everything an agent or human needs to debug why context injection, compression, or sessions behave differently across machines. Install it when wiring AgentMemory during Build, when debugging Operate session issues, or when documenting team defaults for persistent agent memory.3installs11Agentmemory Hooksagentmemory-hooks is a non-user-invocable reference skill that maps how the agentmemory Claude Code plugin registers hooks on twelve lifecycle events so memory lands automatically while you code. Solo builders install the plugin once and rely on session boundaries, tool-use hooks, and prompt hooks to write observations instead of manually invoking memory_save for every task. Use this skill when you need to explain capture behavior to a teammate, debug why an observation never appeared, or decide which events should feed your memory graph. It pairs with the broader agentmemory stack for handoff and continuity across long agent sessions. The skill is meta documentation generated from hooks.json rather than a workflow that mutates your repo by itself.3installs12Agentmemory Mcp Toolsagentmemory-mcp-tools is a dense MCP tool catalog for the agentmemory server, generated from the TypeScript tools registry rather than hand-edited prose. Indie builders and solo agents integrating long-lived memory need accurate parameter lists when choosing between the eight-tool core bundle and the full fifty-three-tool surface. The reference maps each tool to its purpose—creating and updating actions, checkpoints, audit trails, Claude bridge sync, and related memory operations—so you configure --tools core or --tools all deliberately. Treat it as lookup material while building agent workflows or debugging which call blocks an action dependency graph. Regenerate the doc after registry edits to avoid stale signatures. It complements behavioral skills like write-agentmemory-skill, which defines how to author skills, not how to invoke individual MCP methods.3installs13Agentmemory Rest Apiagentmemory-rest-api is a generated REST reference for the AgentMemory local memory server used by solo and indie builders running Claude Code, Cursor, Codex, and similar hosts. It lists all HTTP methods and paths under the default :3111 base, including optional Bearer authentication when AGENTMEMORY_SECRET is set, and documents localhost-open behavior otherwise. Builders use it while implementing automations, custom clients, or CI hooks that call consolidate, checkpoint, cascade-update, auto-forget, and crystal endpoints instead of guessing URLs from source. The skill mirrors skills.sh-style procedural knowledge: read the table, match Method and Path to your task, then implement against the live server. It pairs naturally with agentmemory-agents for MCP wiring but stays at the raw API layer for backends, scripts, and debugging. Because content is autogen from api.ts, it stays accurate when the project evolves—rerun skills:gen after you add triggers.3installs14Write Agentmemory Skillwrite-agentmemory-skill is the canonical style guide for creating or restructuring skills in the agentmemory ecosystem. Solo maintainers and indie builders extending agent memory hit inconsistent SKILL.md files that agents mis-load or confuse with siblings; this skill enforces one skimmable tiered format so discovery, workflows, and cross-links stay predictable. It specifies directory layout, frontmatter fields, description rules, section order, line budgets, and where to park dense tables or transcripts in REFERENCE.md and EXAMPLES.md. Use it when adding a skill, restructuring an existing one, or reviewing a contribution for consistency—not for end-user product features. Matching the house format keeps procedural knowledge accurate, current, and easy for Claude Code, Cursor, or Codex agents to select without bloating the primary SKILL.md.3installs