dancolta/github-repo-audit
GitHub: dancolta/github-repo-audit
Stars: 0 | Forks: 0
# github-repo-audit
[](https://opensource.org/licenses/MIT) [](https://docs.anthropic.com/en/docs/claude-code) [](https://github.com/dancolta/github-repo-audit)
**Audits your GitHub repo's public surface across 5 dimensions, in parallel.**
*For solo OSS maintainers whose working product is buried under a weak README.*
Your README should feel like [ripgrep](https://github.com/BurntSushi/ripgrep)'s or [bat](https://github.com/sharkdp/bat)'s — opinionated, scannable, and confident about what it is and isn't. Not like a SaaS landing page.
## How it works
flowchart LR
A[/github-repo-audit URL/] --> B[Intake: ground truth]
B --> C{5 agents in parallel}
C --> D1[competitive-analyst]
C --> D2[market-researcher]
C --> D3[product-strategist]
C --> D4[ui-ux-designer]
C --> D5[content-marketer]
D1 & D2 & D3 & D4 & D5 --> E[Synthesis]
E --> F[Severity-ranked issue list]
F --> G{Approval gate}
G -->|approved| H[Apply: README + gh repo edit]
G -->|rejected| I[Stop, no writes]
1. **Intake** — you paste a repo URL (or point at a local path) and a one-paragraph "what this product actually is." If you skip that paragraph, the skill drafts one from your README and asks you to confirm.
2. **Five agents in parallel** — each looks at one dimension of your repo's public surface and returns under 800 words.
3. **You approve, the skill applies** — synthesis becomes a severity-ranked issue list (Critical/High/Medium/Low). You pick what to apply. Nothing writes to disk or hits `gh repo edit` without your sign-off.
## Install
mkdir -p ~/.claude/skills && \
git clone https://github.com/dancolta/github-repo-audit ~/.claude/skills/github-repo-audit
That's it. The skill is now available as `/github-repo-audit` (and triggers on phrases like "audit my repo" or "rewrite my readme").
## Usage
/github-repo-audit https://github.com/yourname/yourproject
Then answer the ground-truth question, watch the 5 agents run in parallel, review the issue list, and approve the fixes you want applied.
## What it audits
The 5 dimensions:
## What it produces
- `synthesis.md` — combined findings from all 5 agents
- `issues.md` — Critical/High/Medium/Low issues with proposed fixes
- `README.md` — rewritten in a fixed section order
- `github-metadata.md` — About sentence + topic tags + suggested rename
- `hero-shotlist.md` — frame-by-frame spec for the hero asset
## Companion skill
[`claude-gif`](https://github.com/dancolta/claude-gif) — optional. If installed, the audit offers to render the hero asset shotlist into an actual GIF. The audit works end-to-end without it; you just get the spec instead of the file.
## How it compares
| | github-repo-audit | ChatGPT prompt | README generator | Manual rewrite |
|---|:---:|:---:|:---:|:---:|
| Audits full public surface (name, About, topics, hero, README) | ✓ | ✗ | ✗ | partial |
| Multi-agent: competitive + market + product + UI/UX + SEO | ✓ | ✗ | ✗ | ✗ |
| Severity-ranked issue list before any edit | ✓ | ✗ | ✗ | ✗ |
| Approval gate before applying fixes | ✓ | ✗ | n/a | n/a |
| Hard rejection of AI-slop phrasing (emoji headers, "unleash", filler) | ✓ | ✗ | ✗ | partial |
| Applies fixes via `gh repo edit` (topics, About, name) | ✓ | ✗ | ✗ | partial |
| Preserves maintainer voice | ✓ | ✗ | partial | ✓ |
| Time to actionable diff | ~5 min | ~30 min of prompting | hours of templating | days–weeks |
## FAQ