Assurance and evidence
This page is written from the viewpoint of an independent reviewer asking: what can I verify myself, and what must I take on trust?
What you can verify in ten minutes
git clone https://github.com/shahvatsalofficial/contribsutra-ai && cd contribsutra-ai
npm test # runs the suite; no dependencies to install
node packages/cli/index.mjs audit --out /tmp/audit
sha256sum -c /tmp/audit/SHA256SUMS.txt # on Linux; use Get-FileHash on Windows
npm pack --dry-run # exactly what would be published
npm ls --all # no dependencies
git grep -nE "fetch\(|https?\.request|net\.connect" packages # only loopback requests in the audit check and tests
Evidence index
| Claim | Where to check |
|---|---|
| 104 tests, 100 passed, 0 failed, 4 skipped on win32/x64, Node v24.15.0 | test-results/test-run.tap, test-results/environment.json |
| 3000 + 300 hostile inputs, 0 unexpected crashes | test-results/fuzz-results.json |
| 30 adversarial checks with expected vs observed | samples/audit/audit-report.html |
| Negative controls detect problems | audit G01, G02 and the empty-repository audit test |
| CLI behavior and exit codes | samples/cli-*.txt (real transcripts) |
| Checksums of every proof file | test-results/SHA256SUMS.txt |
| Cross-platform results | GitHub Actions history of the repository (not asserted here) |
| Published package contents | npm view contribsutra-ai and the package page |
Independence and its limits
The audit and the proof are produced by the same repository they assess. That is self-assessment with reproducible evidence. Independence comes from reproducibility: any reviewer can rerun and compare hashes. It does not replace an external security audit, an accessibility audit with assistive technology, or a legal review.
Evidence we deliberately do not show
| Missing evidence | Reason |
|---|---|
| Adoption numbers, stars, downloads | Not measured; would be vanity |
| Testimonials and logos | None exist; none are invented |
| Acceptance rate of contributions made with the tool | Not measured |
| Native activation in Claude Code, Cursor, Codex, Gemini CLI, Copilot | Not yet verified; copying files is verified |
| Live GitHub behavior | The tool does not use the network |
| Results on macOS and Linux from this recording | See CI |
Known limitations
Heuristic policy and license detection can be wrong. Issue scoring is a transparent heuristic, not a prediction. Secret redaction is best-effort. The seal is not a signature. A Windows host without symlink permission cannot execute four symlink tests locally.
How a reviewer should read an audit report
- Check the verdict and the number of failed and skipped checks.
- Read each skipped check's reason.
- Read "Not tested" and decide whether it matters for your use.
- Compare
SHA256SUMS.txtwith your own run. - Treat the report as a statement about this tool on this host, not about your repository's quality.