agent-coach
llms.txt0 starsChanged 3 days agoMITView on GitHub
agent-coach is an open-source, local-first Python CLI that grades and coaches how you prompt Claude Code and OpenAI Codex, using your own conversation history (~/.claude/projects and ~/.codex/sessions) and a rubric built from Anthropic's and OpenAI's official guides and academies. What it does: - Scores 12 prompting habits (specific context, done criteria and verification, constraints, handing over real material, scope, plan first, stop points, asking for evidence, course correction, moving repeated rules into CLAUDE.md/AGENTS.md, one outcome per conversation, matching model and…
- Installs packages
In this repositoryOne other file it gives its agents.
- coachSkill
The file
Files
# agent-coach > agent-coach is an open-source, local-first Python CLI that grades and coaches how you prompt Claude Code and OpenAI Codex, using your own conversation history (~/.claude/projects and ~/.codex/sessions) and a rubric built from Anthropic's and OpenAI's official guides and academies. What it does: - Scores 12 prompting habits (specific context, done criteria and verification, constraints, handing over real material, scope, plan first, stop points, asking for evidence, course correction, moving repeated rules into CLAUDE.md/AGENTS.md, one outcome per conversation, matching model and reasoning effort to the task) on a 0-100 scale, using the model and effort each conversation actually ran on. - Rewrites prompts the user actually typed into better versions, citing the guide section behind each habit. - Finds "missed responses": turns where the user didn't get what they wanted, classified as prompt gap, agent error, changed mind or external limit. Short prompts are not penalized when context carried the intent. - Groups recurring prompt gaps into one-line rules and says where to put them (CLAUDE.md / AGENTS.md, memory, skill, hook). - Saves every run to ~/.agent-coach/history so progress can be tracked over time. - Grades through the user's signed-in `claude` or `codex` CLI (no API key), or offline with keyword heuristics. - Masks secrets and contact details before sending anything to the judge. Zero dependencies, Python 3.10+, MIT license. English and Korean. Install and run: pipx install agent-coach-cli (or uv tool install agent-coach-cli, pip install agent-coach-cli; the command is agent-coach) agent-coach run As a plugin (the agent runs it and coaches in chat): Claude Code: /plugin install agent-coach --marketplace 1017toa/agent-coach, then /agent-coach:coach Codex: codex plugin marketplace add 1017toa/agent-coach && codex plugin add agent-coach@agent-coach ## Docs - [README](README.md): overview, usage, FAQ - [README (Korean)](README.ko.md) - [Rubric](agentcoach/rubric/default.json): the 12 habits with official guide citations - [Contributing](CONTRIBUTING.md)
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

