Assurance and evidence
This page lists the claims DeploySutra makes, how each is verified, and the evidence you can reproduce. Anything not verified is said so.
Claim register
| Claim | Verified by | Recorded result |
|---|---|---|
| Never deploys; no apply/deploy/destroy/rollback command | CLI test for each word; audit scans CLI help | pass |
No network, no subprocess in analysis code, no eval | Audit static scan of packages/ | pass |
| Zero runtime dependencies | package.json check in tests and audit | pass |
| Secrets and plan state never reach reports | Canary strings in env, Secret data, annotations, plan variables, prior state, addresses | 0 found |
| Unsupported YAML fails loudly | 18 parser tests, 15 hostile inputs in audit | all rejected |
| Parser survives mutation | 400 seeded mutations plus 400-level nesting | 0 crashes, 0 hangs |
| Gate can only be BLOCKED or HUMAN_REVIEW_REQUIRED | 50 corpus resources assessed with every rollback box ticked | none ready |
| Equal inputs give equal digests | Determinism check in tests and audit | pass |
| Edited, extended or unsealed reports are detected | Integrity checks | pass |
| Rules do what they say | 50-case corpus | exact (author-labelled) |
| Skills reference real commands and rule IDs | tests/skills.test.mjs | pass |
| MCP tools are read-only and scoped | Scripted client tests, traversal refusal | pass |
| Loopback API needs a token, refuses remote bind and writes | Server tests | pass |
Recorded run
Version 0.2.0 · win32 10.0.26300 · Node v24.15.0 · 2026-11-17T17:09:01.008Z · 308 passed, 0 failed, 4 skipped of 312 · audit 42 passed, 0 failed, 7 not run of 49.
Open samples/audit-report.html for the full audit, samples/review-demo.html for a BLOCKED review, samples/review-hardened-clean.html for a clean review that is still not an approval, and test-results/test-run.tap for the raw log. CLI transcripts are in samples/cli-*.txt.
What was independently reviewed
Nothing. The audit is the tool testing itself, the corpus is written by the author, and no third party has reviewed the code. These documents say so on purpose. If you need independent assurance, review the source (about 2,000 lines, no dependencies) or commission a review; the audit report and corpus give a reviewer a head start.
Not run
Live cluster dry-run, MCP Inspector or third-party MCP clients, native skill activation in each IDE, independent benchmark, signed releases, other operating systems in the recorded local run (CI covers them).
Reproduce everything
git clone https://github.com/shahvatsalofficial/deploysutra-ai && cd deploysutra-ai
npm test
npx deploysutra-ai audit --out audit-out
npx deploysutra-ai verify --report audit-out/audit-report.json
Reading the evidence critically
A pass means "the tool behaved as designed on these inputs". It does not mean "your deployment is safe". Use it as one input to a human decision.