MigraSutra AI Documentation
MigraSutra AI is an offline, deterministic, read-only CLI plus 20 portable agent skills. It audits a repository for migration risk, rehearses a manifest change in a temp directory, and produces a conservative release-gate review as JSON, Markdown and HTML.
Tagline: Upgrade with Evidence. Rehearse Before Change. Recover with Confidence.
Read in this order
| Audience | Document | What you get |
|---|---|---|
| Executives, buyers | Executive summary | Problem, value, honest scope, risk |
| Product, QA | Functional specification | Commands, rules, inputs, outputs, verdicts |
| Engineers | Technical specification | Modules, schemas, algorithms, limits |
| Architects | Architecture | Context, components, data flow, trust boundaries |
| Users | User guide | Install, run, read reports, CI, IDE skills |
| Security, legal | Security, privacy and legal | Threat model, license, claims policy |
| Reviewers | Independent audit and evidence | Method, results, limits, how to reproduce |
| Reviewers | Observed test results | Machine-written run of the real suite |
Sample reports
- Risky fixture: audit-risky.html,
evidence/samples/audit-risky.md,evidence/samples/audit-risky.json - Clean fixture: audit-clean.html,
evidence/samples/audit-clean.md,evidence/samples/audit-clean.json
Three facts to keep in mind
- MigraSutra never executes your tests, installs, scripts or migrations. Test status is always
NOT_RUN. - The best verdict it can issue is
REVIEW_REQUIRED. It never says "safe to deploy". - Its SQL checks are heuristics. They miss dynamic SQL inside dollar-quoted bodies.
The HTML pages in this folder are generated from the Markdown files by npm run build:docs. CI fails if they are stale.