github-repo-candidate-quality-gate — Claude Code skill: what it does
github-repo-candidate-quality-gate turns raw GitHub repo search output into a shortlist with metadata, freshness and provenance labels. Useful when you have to decide which repos are worth recommending.
MIT 254 installs ★ 61 updated 2026-10-24
npx skills add akillness/jeo-skills --skill github-repo-candidate-quality-gate -a claude-codeWhat github-repo-candidate-quality-gate does
This skill takes the noisy output of gh search repos and turns it into a shortlist you can actually recommend. Every repo it keeps gets a metadata record: full_name, url, license, pushed_at (falling back to updated_at), archived, stars, a one-line reason for the fit, and a provenance label.
Gates run in a fixed order: license, archive status, freshness, fit, then traction. A repo with no license is dropped unless someone writes down why it stays. Archived repos go by default. Anything not pushed to in 24 months fails the freshness gate. Broad hits are still counted as discovery evidence, but a keep needs traction, and the default bar is stars >= 3.
Reporting keeps the full discovery counts even when few repos survive. Each lane reports lane_status: pass|degraded plus degraded causes drawn from license, stale, low-fit, archived, low-signal, no-results, alongside kept_count, raw_count, median_stars_raw and zero_star_raw. Raw evidence files are never deleted, so a stricter run can be rolled back. It works on search output you already collected with gh search repos or an equivalent.
What it helps with
- filter out repos with no license from
gh search reposresults - drop archived and stale projects before shortlisting
- keep full discovery counts when few repos survive the gates
- mark a lane degraded and name the causes
- keep raw evidence files intact so a run can be rolled back
Example prompts for the agent
- Run this
gh search reposoutput through the quality gate and keep only the recommendation-grade candidates with metadata. - Build the lane report with
lane_status,degraded_causesandkept_count. Don't delete the raw files. - Drop archived and stale repos, and write the risk note for any exception keep.
How to install github-repo-candidate-quality-gate
1Skills CLI (Vercel)
npx skills add akillness/jeo-skills --skill github-repo-candidate-quality-gate -a claude-code2Manually: put the files in place
~/.claude/skills/github-repo-candidate-quality-gate/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/github-repo-candidate-quality-gate/
1Skills CLI (Vercel)
npx skills add akillness/jeo-skills --skill github-repo-candidate-quality-gate -a codex2Manually: put the files in place
.agents/skills/github-repo-candidate-quality-gate/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add akillness/jeo-skills --skill github-repo-candidate-quality-gate -a cursor2Manually: put the files in place
.cursor/skills/github-repo-candidate-quality-gate/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install akillness/jeo-skills .agent-skills/github-repo-candidate-quality-gate/SKILL.md2Manually: put the files in place
.github/skills/github-repo-candidate-quality-gate/SKILL.md1Skills CLI (Vercel)
npx skills add akillness/jeo-skills --skill github-repo-candidate-quality-gate -a gemini-cli2Platform CLI
gemini skills install https://github.com/akillness/jeo-skills --path .agent-skills/github-repo-candidate-quality-gateManually: put the files in place
.devin/skills/github-repo-candidate-quality-gate/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add akillness/jeo-skills --skill github-repo-candidate-quality-gate -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about github-repo-candidate-quality-gate
How do I install github-repo-candidate-quality-gate in Claude Code?
Run npx skills add akillness/jeo-skills --skill github-repo-candidate-quality-gate -a claude-code. The skill file lands at ~/.claude/skills/github-repo-candidate-quality-gate/SKILL.md.
What does github-repo-candidate-quality-gate do?
It filters GitHub search results through license, archive, freshness, fit and traction gates, then writes a per-lane report with the reasons a lane came out degraded.
Which tools does the skill use?
allowed-tools lists Bash, Read, Write, Grep and Glob, which covers reading the search output, grepping through it and writing the report.
Source text
SKILL.md · MIT
--- name: github-repo-candidate-quality-gate description: > Convert noisy GitHub repository search results into recommendation-grade candidate lists with explicit metadata, freshness, traction signal, provenance labels, and rollback-safe reporting for maintenance PR workflows. allowed-tools: Bash Read Write Grep Glob license: MIT metadata: tags: github, search, triage, quality-gate, repo-maintenance version: "1.0.0" source: akillness/jeo-skills --- # GitHub Repo Candidate Quality Gate Use this skill when broad GitHub search output is too noisy for merge-ready recommendations. ## When to use - You already ran `gh search repos` (or equivalent) and need recommendation-grade keeps. - You must preserve broad discovery evidence while filtering keeps deterministically. - You need lane-level status + degraded-cause reporting for unattended cron review. ## Minimum keep metadata For every kept candidate, capture all fields: - `full_name` - `url` - `license` - `pushed_at` (or `updated_at` fallback) - `archived` - `stars` - one-line fit rationale - provenance label ## Quality gate order 1. **License gate**: drop missing/unknown license unless explicit exception rationale. 2. **Archive gate**: drop archived repos by default. 3. **Freshness gate**: drop if `pushed_at` older than 24 months unless exception + risk. 4. **Fit gate**: drop keyword-only mismatches (low-fit). 5. **Signal gate**: keep broad discovery evidence, but recommendation-grade keeps require traction (default: `stars >= 3`) or explicit adoption rationale. ## Reporting contract - Keep and publish full discovery counts even when keeps are sparse. - Emit `lane_status: pass|degraded`. - Emit `degraded_causes` counts using: `license`, `stale`, `low-fit`, `archived`, `low-signal`, `no-results`. - When a lane ends with `raw_count == 0` after documented recovery, include `no-results` explicitly to keep degraded reports deterministic. - Include lane health metrics: `kept_count`, `raw_count`, `median_stars_raw`, `zero_star_raw`. - State provenance labels used (`direct page retrieval`, `indexed snippet`, etc.). ## Rollback-safe operation - Never delete raw evidence files. - If stricter gates reduce keep counts too far, preserve output but mark lane degraded. - Allow exception keeps only when rationale + risk note are explicit. ## Example - Input: 30 raw hits, many zero-star repos. - Output: discovery evidence retained, recommendation keeps filtered by metadata/freshness/signal gates, degraded causes documented for reviewer.
© akillness · MIT · akillness/jeo-skills