Test report
Recorded by npm run proof for version 0.2.0 on win32 10.0.26300, Node v24.15.0, at 2026-11-17T17:09:01.008Z. Raw TAP log: test-results/test-run.tap.
Summary
| Measure | Result |
|---|---|
| Tests | 312 |
| Passed | 308 |
| Failed | 0 |
| Skipped | 4 (symlink tests: this OS user cannot create symlinks; CI runners can, and run them) |
| Rule corpus | 50/50 exact |
| Self-audit | 42 passed, 0 failed, 7 not run (listed) |
What the suites cover
| File | Focus |
|---|---|
tests/core.test.mjs | Hashing and sealing, YAML parser (accept, reject, source maps), ingestion limits, diff and redaction, rules, assess and gate, BlastMap, rollback, plan import, report rendering and escaping |
tests/integration.test.mjs | CLI exit codes and outputs, --fail-on, profiles, multi-document, audit command, skill installer for six IDE layouts, loopback server auth/methods |
tests/mcp.test.mjs | JSON-RPC protocol, tool list, traversal refusal, no state in plan summaries, tamper detection |
tests/skills.test.mjs | 25 skills: names, sections, real commands, real rule IDs; 20 agents; package consistency; repository-wide secret-shape scan |
tests/audit.test.mjs | The self-audit itself: sections, no failures, no certification claim |
Negative and security tests (selection)
Anchors, aliases, tags, templates, tabs, block scalars, merge keys, duplicate and __proto__ keys, odd indentation, unterminated quotes; files over 1 MiB, wrong extension, missing file, directory, symlink file and directory, more than 200 documents, duplicate identity; plan with variables/prior_state/addresses/values; hostile HTML and Markdown in annotations; edited, extended and unsealed reports; weak token, remote bind, POST/PUT/DELETE/PATCH, wrong path; absolute paths, .. and NUL bytes through MCP.
Self-audit highlights
The audit plants canary strings in env values, Secret data, annotations, plan variables, prior state and addresses, then searches every output: none may appear. It feeds 15 hostile YAML inputs and 400 seeded mutations to the parser, checks determinism and tamper detection, and verifies that no corpus resource can produce any gate except BLOCKED or HUMAN_REVIEW_REQUIRED even with all rollback boxes ticked.
Performance (median of 5)
| Operation | Result |
|---|---|
assess() over 200 resources | 37 ms |
| Parse 200 documents | 12 ms |
| Rules over 200 resources | 1 ms |
Not tested
Real clusters, kubectl diff, Helm/Kustomize/OpenTofu binaries, MCP Inspector, native skill activation in each IDE, signed releases, independent rule benchmark. macOS and the full Node 20/22/24 matrix are covered by CI, not by this recorded local run.
Reproduce
git clone https://github.com/shahvatsalofficial/deploysutra-ai && cd deploysutra-ai
npm test
npm run proof # rewrites docs/proof