Executive summary
ContribSutra AI helps a first-time open-source contributor become a respectful, capable contributor, and helps maintainers receive fewer, better, more reviewable contributions. Version 0.2.1 is a working, offline command-line tool plus 25 portable agent skills.
The problem
Beginners struggle to learn what a repository expects, whether their machine can run it, which issue is safe to pick and what "ready for review" means. Maintainers pay for that confusion with low-quality, duplicate and unwanted pull requests, now amplified by AI tools that generate changes at volume.
The product
| Capability | What the user gets |
|---|---|
| Repository guide | Policies, license, toolchain, CI and contribution rules, each with file:line and a SHA-256 |
| Setup doctor | Which tools are installed and whether they satisfy the versions the repository declares |
| First-win radar | A ranked issue shortlist with reasons, and clear refusals for risky issues |
| Learning path | Six small steps with hints and acceptance checks, in three learning modes |
| PR ProofDesk | A review packet that separates verified, user-reported, inferred and not-run claims |
| Independent audit | An HTML, Markdown and JSON report with 30 adversarial checks and a negative control |
Why it can be trusted
- It cannot do harm: no GitHub writes, no network, no execution of repository scripts. There is no code path that opens a pull request.
- It does not exaggerate: a test claim supplied by a person stays
unverified_external_claim. The tool never emits "approved". - It shows its work: the recorded run executed 104 tests (100 passed, 0 failed, 4 skipped on this host), plus 3000 hostile issue snapshots and 300 hostile document sets with 0 unexpected crashes.
- It is honest about limits: native activation of the skills in each IDE, live GitHub state and adoption outcomes are listed as not tested or not measured.
Who it is for
First-time contributors, students and bootcamp graduates, engineers entering unfamiliar repositories, maintainers who want lower support burden, educators, developer-relations teams and organizations running onboarding events.
Business and community value
Lower time to a first useful contribution without rewarding spam; less maintainer time spent on triage; reproducible onboarding material that can be reviewed by a person. No outcome metric (acceptance rate, retention, satisfaction) is claimed: none has been measured yet.
Status at a glance
| Item | State |
|---|---|
| Version | 0.2.1 |
| License | Apache-2.0 |
| Runtime dependencies | 0 |
| Network access | None in the CLI |
| Third-party certification | None. The audit is a reproducible self-assessment |
| Next milestone | Read-only GitHub adapter and git change summary (see roadmap) |