|
Florian BRUNIAUX · AI Founding Engineer @ Méthode Aristote 13 years from developer to CTO / VP Eng · Blog ↗ · Projects ↗ |
A Claude Code plugin that audits a GitHub profile through its author's intent. It reviews READMEs, flags AI-generated markers, reads recruiter signals, and suggests keywords for findability. Born from The Product Crew (TPC) GitHub Roast live, ep. 2.
Each repo is read against its objective, not against code quality in the abstract. Four objectives form the grid: build a flagship, raise funds, get hired, run a business. Few stars does not mean low value.
Technical principle: orchestrate, do not rewrite. Existing building blocks (organic score and geo on the StarMapper side, anti-AI detection on the personal skills side) are called, not duplicated.
| Phase | Content | Status |
|---|---|---|
| 0 | Plugin scaffold | done |
| 1 | eval-readme (skill + readme-critic agent) | done |
| 2 | analyze-github-profile (objective, pins, timeline, commits) | done |
| 3 | score-profile (/100 grade + table + fixes + score card) | done |
| 4a | suggest-keywords (findability, SEO, cross-linking) | done |
| 4b | analyze-linkedin-profile (multimodal input) | done |
| 5 | route-next-steps (TPC offers / Reddit post draft) | done |
| 6 | community packaging | upcoming |
commands/: six slash commands (gh-readme,gh-profile,gh-score,gh-keywords,gh-linkedin,gh-next)skills/: six skills holding the procedures and gridsagents/: four agents (readme-critic,github-profile-analyst,keyword-strategist,linkedin-analyst)signals.md: shared signal grid, the single source of truth
- Claude Code (the plugin runs inside it, there is nothing to deploy or host).
- The GitHub CLI
gh, authenticated (gh auth login). It is the core data source: profile, repos, pinned items, contribution timeline, commits, issues and PRs. - Optional: a StarMapper instance for organic score and stargazer geography. The skills call its public HTTP API when available and fall back to raw
ghsignals otherwise. Nothing to run locally.
This repo is a single-plugin Claude Code marketplace. From inside Claude Code:
/plugin marketplace add FlorianBruniaux/github-roast-tpc
/plugin install github-roast-tpc@github-roast-tpc
The first command registers the marketplace (it reads .claude-plugin/marketplace.json), the second installs the plugin. Restart Claude Code if the slash commands do not show up immediately.
Local development, to test changes before pushing:
/plugin marketplace add /absolute/path/to/github-roast-tpc
/plugin install github-roast-tpc@github-roast-tpc
Verify it loaded with /help: the gh-readme, gh-profile, gh-score, gh-keywords, gh-linkedin and gh-next commands should be listed.
Two ways to trigger the plugin once installed.
Slash commands drive a specific stage on demand, for example /gh-profile USERNAME or /gh-readme owner/repo (full list below). Each command loads its skill and the matching agent.
Skills also fire automatically from natural language. Asking "audit my github" or "is this README AI-generated?" loads the relevant skill without typing a command. For a full end-to-end run, use the one-shot meta-prompt in the section further down.
Review a single README:
/gh-readme owner/repo
The agent fetches the README via the GitHub API, applies the eval-readme grid, and returns a human/AI verdict, strengths, and ready-to-apply fixes.
Analyze a full profile:
/gh-profile username
The agent fetches repos, pinned items, timeline and commits, detects the author's main objective, and returns a profile sheet with signals by category and points of attention.
Score a profile out of 100 with fixes and a shareable card:
/gh-score username
Chains profile analysis, flagship README review, then a /100 grade weighted by the detected objective, a signal table, fixes sorted by return on effort, and a score card ready to screenshot or download as PNG.
Optimize findability:
/gh-keywords username
Suggests keyword-rich descriptions and topics per repo, an optimized bio, and the missing cross-platform links. Everything is copy-paste ready into GitHub.
Cross-check your LinkedIn against your GitHub:
/gh-linkedin username
You provide your own profile (screenshots or pasted text), the agent checks the alignment with your GitHub and returns fixes. No scraping, no storage.
Plan next steps:
/gh-next username
Branches on your objective. Job track: a fix checklist then a bridge to TPC offers. Outreach track: a Reddit or LinkedIn post draft ready to publish.
Run the whole pipeline in a single shot. Paste this into a Claude Code session that has the plugin loaded, replacing USERNAME and OWNER/REPO:
Run a complete github-roast-tpc audit for USERNAME (flagship: OWNER/REPO).
Go through every stage in order and produce one consolidated report:
1. analyze-github-profile: detect the main objective, then run the interactive
intake: reflect your hypothesis back to me and ask up to 3 targeted questions.
Wait for my answer before continuing.
2. eval-readme on the flagship: human vs AI verdict, fixes.
3. score-profile: /100 grade weighted by the confirmed objective, signal table, and score card generated locally.
4. suggest-keywords: keyword-rich descriptions, topics, bio, llms.txt / SEO gaps.
5. route-next-steps: next actions for the confirmed objective.
Use the gh CLI as the data source. Cite the gh command behind every number.
Stop at stage 1 for my confirmation, then run 2 to 5 without further questions.
The intake pause in stage 1 is intentional: the objective drives the weighting of every later stage. Everything after the confirmation runs end to end.
Once the plugin is installed, three things to configure. Only the first is mandatory.
- GitHub CLI. Authenticate
ghso the skills can read profiles, repos, timelines, commits, issues and PRs:
gh auth login
gh auth status # confirm you are logged in
A token with the default scopes is enough for public data. Without auth, the GitHub API caps at 60 requests/hour and most calls fail.
- TPC job offers link, used by
/gh-nexton the job track. It is not hardcoded. Provide it at runtime when the command asks, or set it here once:
TPC_OFFERS_URL = (to fill in)
- StarMapper integration (optional). It powers the organic score and stargazer geography. No environment variable and no local server are needed: the skills call the public StarMapper HTTP API (
https://starmapper.bruniaux.com/api/mcp/...) when a repo has been scanned, and fall back to rawghsignals (forks, watchers, releases) otherwise. To use a self-hosted StarMapper, give its base URL to the skill when prompted. Reminder: the organic score is gated at 500 stars, below which it returnsinsufficient.
These projects extend the workflow without duplicating this tool:
- Visualize with StarMapper: add geographic audience analysis through the existing optional integration.
- Learn with Claude Code Ultimate Guide: connect the plugin to Claude Code review and extension practices.