memex-agentic-memory
A-MEM-inspired agentic memory workflow for structured knowledge capture.
Audit and sync agent instruction files across all coding agent formats. FRE (first-run) checks scaffolding completeness; ongoing use keeps files in sync after edits.
$ npx -y skills add iamtouchskyer/memex --skill agent-prompts-warmup --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/agent-prompts-warmupContext preview
The summary Claude sees to decide when to auto-load this skill.
Audit and sync agent instruction files across all coding agent formats. FRE (first-run) checks scaffolding completeness; ongoing use keeps files in sync after edits.
name: agent-prompts-warmup description: "Audit and sync agent instruction files across all coding agent formats. FRE (first-run) checks scaffolding completeness; ongoing use keeps files in sync after edits." triggers: - "agent prompts warmup" - "warmup agent prompts" - "audit agent docs" - "check agent instructions" - "update AGENTS.md" - "add agent target" - "sync agent docs"
Agent instruction files ensure every coding agent (Claude Code, Codex, Copilot, Cursor, Gemini, Windsurf/Anti-Gravity) reads the same project rules when working on this codebase.
**Single source of truth**: `AGENTS.md`
All other files are **copies**, synced automatically:
| File | Agent(s) | |------|----------| | `AGENTS.md` | Codex CLI, GitHub Copilot (source of truth) | | `CLAUDE.md` | Claude Code | | `GEMINI.md` | Gemini CLI, Jules, Anti-Gravity | | `.cursorrules` | Cursor | | `.windsurfrules` | Windsurf / Anti-Gravity (legacy) |
**Sync mechanisms** (4 layers): 1. `scripts/postbuild.mjs` — `npm run build` copies AGENTS.md → all targets 2. `scripts/pre-commit` — git hook blocks commit if any file has drifted 3. `tests/integration/agent-docs-consistency.test.ts` — `npm test` fails on drift 4. CI — GitHub Actions runs `npm test` on push/PR to main
Run a full scaffolding check:
/agent-prompts-warmup audit
**Checklist to verify:**
**If anything fails**: fix it immediately. The fix is usually `npm run build` (syncs files) or `npm run prepare` (installs hook).
When the user edits `AGENTS.md` (or asks you to):
1. **Edit only `AGENTS.md`** — never edit the copies directly 2. Run sync: `node scripts/postbuild.mjs` (or `npm run build`) 3. Verify: `npm test -- tests/integration/agent-docs-consistency.test.ts` 4. Confirm all 5 files are identical
When a new coding agent emerges that reads a different file:
1. **Identify the file name** the agent reads (e.g., `.devinrules`, `CLINE.md`) 2. Update `scripts/postbuild.mjs` — add `cpSync("AGENTS.md", "<new-file>");` 3. Update `tests/integration/agent-docs-consistency.test.ts` — add `expect(readFileSync("<new-file>", "utf-8")).toBe(agents);` 4. Update `scripts/pre-commit` — add the new file to the `TARGETS` array 5. Create the file: `cp AGENTS.md <new-file>` 6. If it should be npm-published: add to `package.json` `files` array 7. Run `npm test` to verify 8. Update this skill's mapping table above
**"ERROR: CLAUDE.md has drifted from AGENTS.md"** → Someone edited a copy instead of AGENTS.md. Fix: `cp AGENTS.md CLAUDE.md` (or whichever file drifted). Then check if the edit in the copy should be applied — if so, edit AGENTS.md first, then `npm run build`.
**Pre-commit hook not running** → `npm run prepare` to reinstall. Or manually: `cp scripts/pre-commit .git/hooks/pre-commit && chmod +x .git/hooks/pre-commit`
**CI failing on agent docs** → `npm run build && npm test` locally. If it passes locally but fails in CI, check that the build step runs before test in the workflow.
// scripts/postbuild.mjs
cpSync("AGENTS.md", "CLAUDE.md"); // Claude Code
cpSync("AGENTS.md", "GEMINI.md"); // Gemini CLI / Jules / Anti-Gravity
cpSync("AGENTS.md", ".cursorrules"); // Cursor
cpSync("AGENTS.md", ".windsurfrules"); // Windsurf / Anti-Gravity (legacy)Persistent memory for AI coding agents. Your agent remembers what it learned across sessions.
A-MEM-inspired agentic memory workflow for structured knowledge capture.
Zettelkasten best practices for building a high-quality knowledge graph.
Load prior knowledge from Zettelkasten memory when the task likely benefits from past context.