Agent skill
ainotes
This skill should be used when the user asks to "consolidate notes", "summarize ainotes", "clean up notes", "ainotes", or wants to consolidate accumulated agent observations into a compact summary.
Install this agent skill to your Project
npx add-skill https://github.com/ianphil/my-skills/tree/main/ainotes
SKILL.md
AI Notes — Consolidate
Consolidate raw session observations from .ainotes/log.md into a curated .ainotes/memory.md.
Workflow
- Check if
.ainotes/directory exists in the repo root. If not, create it with emptymemory.md,rules.md, andlog.md. - Read
.ainotes/log.md— these are raw observations appended by the commit skill. - Read
.ainotes/memory.md— this is the current curated memory. - Read
.ainotes/rules.md— check for any rules that should be updated or deduplicated. - Consolidate log entries into memory:
- Merge new observations into the appropriate section
- Deduplicate — if a fact is already in memory, skip it
- Merge related observations into single concise entries
- Prune stale info that contradicts newer observations
- Keep memory under ~200 lines (hard limit)
- If any log entries describe mistakes or operational lessons, ensure they're captured in
rules.mdas one-liners. - Write updated
memory.md - Truncate
log.md— keep only the last 10 entries as a buffer, remove everything else.
memory.md Format
Use structured sections. Only include sections that have content:
# AI Notes — <project name>
## Architecture
- <observation>
## Gotchas
- <observation>
## Workflows
- <observation>
## Testing
- <observation>
## Dependencies
- <observation>
## Conventions
- <observation>
Rules
- Every bullet must be a terse one-liner — no paragraphs
- Prefer specifics over generalities (file paths, command names, exact behavior)
- If memory exceeds ~200 lines, aggressively prune least-useful entries
- Never duplicate information already in README.md or AGENTS.md
- When merging contradictory observations, keep the newer one
- Mistakes and operational lessons go in
rules.md, notmemory.md
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
commit
This skill should be used when the user asks to "commit changes", "push my code", "commit and push", "save my work", or wants to stage all changes and push to remote.
prime-feat
This skill should be used when the user asks to "load feature context", "resume feature work", "prime feature N", "continue feature N", or wants to load all planning artifacts for a specific feature number.
implement-agents
This skill should be used when the user asks to "implement in parallel", "run phases concurrently", "parallel implement", "implement-agents phase X phase Y", or wants to orchestrate multiple agents running /implement simultaneously.
code-walkthrough
Generate structured, hallucination-proof code walkthroughs using Showboat. Use when the user asks to "walk me through this code", "explain this codebase", "create a walkthrough", "document how this works", "give me a tour of this repo", or wants a readable narrative document that explains code with real verified snippets. Also triggers on "showboat", "linear walkthrough", or "code documentation". Not for quick code questions — this produces a full markdown document.
implement
This skill should be used when the user asks to "implement phase N", "implement task T001", "do TDD for phase", "implement T001-T009", or wants to execute tasks following strict TDD red-green-refactor workflow.
work-plan
This skill should be used when the user asks to "show work plan", "what's the plan for feature N", "show progress", "view execution plan", or wants to see phase breakdown, progress status, and next actions for a feature.
Didn't find tool you were looking for?