Auto improve
Skill pablodiazjorge/prompt-forge/.github/skills/auto-improve
Drop-in toolkit that gives AI coding agents persistent memory across sessions. Agent Skills, cross-session learning loop, and knowledge persistence. Zero dependencies, no backend.
npx -y skills add pablodiazjorge/prompt-forge --skill auto-improveAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Post-session learning loop. Use at the END of every coding iteration or when the user says "analyze this session", "what did we learn", "save lessons", or "run auto-improve". Reviews the chat for errors, fixes, API discoveries, and patterns. Cross-references with the issue registry to detect recurrence. Promotes confirmed patterns to user memory, repo memory, or new skills. NEVER invoke mid-task — only at natural stopping points.
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
2.1 KB, as published. Nobody here has run it
Auto-Improve (Legacy Wrapper)
⚠️ This skill has been split into two specialized roles. It is retained for backward compatibility with existing references. New sessions should use the individual skills directly.
What Changed
The original auto-improve skill combined two distinct responsibilities into
one monolithic workflow. It has been split into:
| Role | Skill | When Loaded | What It Does |
|---|---|---|---|
| Developer | developer/SKILL.md | By default (always) | Phases 1-3: scan chat → create/update issues in open/. NEVER promotes or touches skills. |
| AI Engineer | ai-engineer/SKILL.md | On demand (subagent) | Phases 4-6 + dedup + recategorization: review ALL issues → deduplicate → recategorize → promote to skills/memories → clean stale entries. |
How to Use
-
Everyday tracking: The
developerskill is loaded by default. It runs at the end of every iteration, recording issues inknowledge/issues/open/. You don't need to do anything — it works automatically. -
Knowledge cleanup: When you want to organize accumulated issues, run the
ai-engineerskill as a subagent. It will review all issues, deduplicate, recategorize, and promote confirmed patterns.
If you're reading this because an old reference triggered auto-improve, the
agent will automatically redirect to developer for the current session's
tracking needs. For promotion and cleanup, invoke ai-engineer explicitly.