| name | agents-memory |
| description | Persistent project memory bootstrap and maintenance. Use in any repository with (or needing) an `AGENTS.md` + `./agents/` system. **Triggers:**\n* **Initialization:** Starting in a new repo, running `/init`, or requests to "set up agents memory."\n* **Development:** Planning features, proposing architecture (→ decisions), writing code (→ conventions), or navigating the codebase (→ architecture).\n* **Debugging:** Fixing complex bugs, investigating issues, or encountering recurring mistakes (→ known-errors).\n* **Memory Management:** Fetching or proposing workflows, glossary terms, conventions, decisions, or known errors. |
Agents Memory
Persistent memory system for coding projects. It creates one router file (AGENTS.md) plus topical memory files under .agents/ so agents can load only the project context they need.
Use When
- The user runs
/init or asks to initialize project memory.
- Starting work in a repo that lacks
AGENTS.md or .agents/.
- The user asks to set up, inspect, or update agent memory files.
- A repeated project convention, decision, workflow, glossary term, or known error should be proposed for memory.
Files
AGENTS.md
.agents/
architecture.md
conventions.md
decisions.md
glossary.md
workflow.md
known-errors.md
Bootstrap Workflow
- Inspect the current project before writing memory files: root files, package/dependency files, existing docs, and existing
AGENTS.md or CLAUDE.md.
- If project memory already exists, update only missing files or clearly stale references after confirming the intent.
- Create
AGENTS.md from templates/AGENTS.md and .agents/* from the matching templates.
- Replace placeholders with facts discovered in the repo. Leave sections as
None yet. when there is no evidence.
- Do not invent conventions, decisions, commands, or architecture.
- Summarize created or updated memory files and any sections that still need user input.
Operating Rules
- Read
AGENTS.md first in projects that have it.
- Lazy-load
.agents/* by need; do not read all memory files by default.
- Treat closed decisions in
.agents/decisions.md as authoritative unless the user explicitly reopens them.
- If code contradicts memory, trust the current code and flag the memory entry as stale.
- Memory updates during normal work are propose-only unless the user explicitly asks to write them.
Trigger Map
| Situation | Read or update |
|---|
| Navigating repo or changing module boundaries | .agents/architecture.md |
| Writing code | .agents/conventions.md |
| Refactor, dependency swap, architecture choice | .agents/decisions.md |
| Project-specific acronym, domain term, entity | .agents/glossary.md |
| Build, test, release, commit, PR | .agents/workflow.md |
| Debugging a repeated failure | .agents/known-errors.md |
Templates
Templates live in templates/ next to this skill. Use them as structure, not as final content.