Roadmap and FAQ
Roadmap
| Phase | Scope | State |
|---|---|---|
| 0 | Truthful baseline: inspect, test, fix regressions | Done |
| 1 | Robust no-key core: scan with line sources, doctor with requirement comparison, adversarial issue handling, audit pack | Done in 0.2.1 |
| 2 | Developer environment: safe git change summary, Dev Container checks, opt-in sandboxed test runner with captured output | Next |
| 3 | GitHub read-only: paginated issues, ETags, rate limits, staleness display | Planned |
| 4 | Quality and learning: richer opt-in explanations, patch review with attributable evidence | Planned |
| 5 | Multi-IDE evidence: real activation tests in each host | Planned |
| 6 | Server and dashboard: OIDC, RBAC, tenant isolation, accessible UI | Planned |
| 7 | Controlled writes: draft PRs only, behind four gates | Planned, optional |
| 8 | Open-source release hardening: signed releases, SBOM, external review | Ongoing |
The four gates for any future write: an explicit user request, a minimum-scope token, a maintainer-aligned invitation and a fresh human review of the diff and evidence. Even then the result is a draft, never a merge.
FAQ
Is this a bot or an AI that writes my pull request? No. It prepares you and your reviewer. You write the change and decide whether to submit.
Do I need an API key, an account or internet access? No. Everything runs offline on Node.js 20 or newer.
Why was a "good first issue" declined? Labels go stale. The issue may be closed, assigned, locked, security-related, very large or contain instruction-like text. The reason is printed next to it.
Why does my passing test show as "unverified"? ContribSutra did not run it. The label tells the reviewer what to trust. Attach the captured output.
Does it work in my IDE? The skill files are copied and tested. Whether your IDE activates them has not been verified. The CLI works everywhere Node.js does.
Can maintainers use it? Yes. scan and audit show what newcomers can see: missing onboarding files, license, policy signals and CI.
Is the audit independent? It is a reproducible self-assessment generated by the tool it audits. You can rerun it and compare hashes. It is not third-party certification, and every report lists what was not tested.
What does "skipped" mean in the test report? The host could not perform the test (for example creating symlinks on Windows). Skips are never counted as passes.
How is it different from issue-discovery sites? It explains suitability, declines risky issues, extracts project policy and keeps evidence honest, offline. See docs/COMPETITOR_ANALYSIS.md for the limits of that comparison.
Who made it? Vatsal Shah. Apache-2.0.