openkata-eval-runner
Skill1 starsChanged 16 days agoMITView on GitHub
Run eval scenarios for an OpenKata skill locally and iterate fixes until 95% threshold. Use when the user says "run evals", "check evals", "eval this skill", or when validating skill quality.
In this repository20 other files it gives its agents.
- AGENTS.mdAGENTS.md
- openkata-optimize-skillsSkill
- openkata-review-ruleSkill
- openkata-review-skillSkill
- openkata-rule-conventionsSkill
- openkata-ryu-releaseSkill
- openkata-skill-conventionsSkill
- commit-conventionsSkill
- create-adrSkill
- create-evalsSkill
- create-profileSkill
- create-ruleSkill
- create-skillSkill
- critical-thinkingSkill
- github-create-prSkill
- git-worktreeSkill
- grill-meSkill
- grill-with-docsSkill
- makefile-conventionsSkill
- spec-workflowSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Run eval scenarios locally and iterate skill fixes until the 95% threshold is met.
- Very specific: line 42
If failing, report which scenarios/criteria failed with reasons, and suggest specific SKILL.md fixes.
- Distinctive: line 69
Does NOT publish, tag, or release
- Distinctive: line 70
Does NOT use tessl eval run for scoring
- Very specific: line 74
Only distributable skills (skills/) have evals
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Eval Runner
Run eval scenarios locally and iterate skill fixes until the 95% threshold is met.
Tool Split
| Tool | Use for |
./bin/openkata-eval | Scoring (runs locally, same model as users) |
tessl skill review --optimize | Instruction polish after passing |
tessl eval run | NOT used — model mismatch makes results unreliable |
Workflow
- Identify the skill — Determine which skill from the
user's request. Resolve path (skills/
<name>/). - Check evals exist — Verify
skills/<name>/evals/has scenario directories. If missing, activate thecreate-evalsskill first. - Build the runner —
make eval-localbuilds the binary automatically if missing. - Run evals — Execute:
The overall average must be 95% or above../bin/openkata-eval skills/<name> - Report — If passing (95%+), report the score. If failing, report which scenarios/criteria failed with reasons, and suggest specific SKILL.md fixes.
- Fix and retry — If the user confirms fixes, apply them to SKILL.md, then re-run. Iterate until 95%+.
- Optimize — Once passing locally, run:
Apply any worthwhile suggestions, re-run local evals to confirm no regression.tessl skill review --optimize skills/<name> - Commit — Stage changes and commit.
Single Scenario Debugging
To iterate on one failing scenario without running all:
./bin/openkata-eval skills/<name>/evals/scenario-XSingle scenario mode always exits 0 (debug tool).
Boundaries
- DOES run evals, suggest and apply SKILL.md fixes
- DOES run tessl optimize for polish after passing
- Does NOT publish, tag, or release
- Does NOT use tessl eval run for scoring
Gotchas
- Only distributable skills (
skills/) have evals - Conversational skills need
"sandbox": falsein scenario.json (no Docker required) - Skills that use tools need
"sandbox": true(default) and Docker running
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.