Evidence
Generated by node scripts/build-evidence.mjs on Node v24.15.0 (win32 x64). Nothing here is hand-written.
Test run
| Tests | Passed | Failed | Skipped | Cancelled | Exit code |
|---|---|---|---|---|---|
| 54 | 52 | 0 | 2 | 0 | 0 |
Full output: test-run.txt. Machine-readable: test-summary.json. Skipped tests are POSIX-only symlink checks that cannot run on Windows. Linux and macOS results come from public GitHub Actions runs, not from this file.
Exit codes behave as documented
| Command | Exit code | Documented | Matches |
|---|---|---|---|
demo | 0 | 0 | yes |
audit | 0 | 0 | yes |
review (demo capsule) | 0 | 0 | yes |
verify (unchanged demo) | 0 | 0 | yes |
verify (after ADR edit) | 2 | 2 | yes |
diff | 0 | 0 | yes |
Redaction and injection proof
A fixture repository held a fake password line, a .env file and a line telling an AI to ignore its instructions. Result: none of the three appears in the capsule or the rendered context. See fixture-capsule.json. The password is a fake value built in the script; no real credential was used.
Sample outputs
- Demo capsule: capsule.json, claude-context.md, report.md, report.html
- Privacy gate: review.json
- Decisions: decisions.json; tasks: tasks.json
- Freshness: verify-current.json (CURRENT), verify-stale.json (STALE after an ADR status edit)
- Drift diff: diff-after-adr-change.json
- Self-audit: audit.md, audit.html, audit.json
Honest limits
- Inputs are a small synthetic repository. They show behaviour, not performance on your codebase.
- The self-audit runs in the same process as the code it checks. It is an integrity check, not a third-party attestation.
- No live IDE, model or cloud service was used to produce any file here.