- Home
- Skills
- Mitsuhiko
- Agent Stuff
- Improve Skill
improve-skill_skill
- TypeScript
705
GitHub Stars
1
Bundled Files
2 months ago
Catalog Refreshed
4 months ago
First Indexed
Readme & install
Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.
Installation
Preview and clipboard use veilstrat where the catalogue uses aiagentskills.
npx veilstrat add skill mitsuhiko/agent-stuff --skill improve-skill- SKILL.md5.7 KB
Overview
This skill analyzes coding agent session transcripts to improve an existing skill or to create a new reusable skill based on what happened in a session. It supports transcripts from Claude Code, Pi, and Codex and provides a repeatable prompt workflow to produce targeted improvements or new skill drafts. Use it to capture lessons, examples, and fixes surfaced during real agent runs.
How this skill works
The skill extracts a session transcript using a small helper script and then generates a structured prompt to run in a fresh agent session. The prompt asks the new session to compare the transcript with the current skill or to distill a new skill, focusing on missing instructions, examples, pitfalls, and concrete fixes. Outputs are written back to the chosen skill location for review.
When to use it
- After a debugging or coding session where the agent struggled or improvised
- When you want to formalize a successful ad-hoc workflow into a reusable skill
- When documentation is missing key examples or step-by-step commands
- To capture recurrent errors and their resolutions into guidance
- Before sharing a skill so it's concise and reproducible for other users
Best practices
- Extract the complete session transcript to preserve context and edge cases
- Run the generated prompt in a fresh agent session to avoid context leakage
- Highlight concrete commands, code snippets, and file paths discovered during the run
- Focus improvements on clarity: add missing examples and remove redundant text
- Record common failure modes and short remediation steps
Example use cases
- Improve an existing coding helper after multiple retries revealed missing options
- Create a new skill that automates a repeated refactor pattern shown in a transcript
- Capture CLI commands and flags an agent discovered while configuring a tool
- Document a workaround that resolved an environment-specific error
- Turn a long interactive debugging session into a short reproducible recipe
FAQ
Supports session transcripts from Claude Code, Pi, and Codex as produced by the extraction script.
Why run the improvement in a fresh agent session?
A fresh session avoids prior context inflation, keeps the prompt focused, and reduces token usage so the agent can analyze the transcript cleanly.