Roadmap and FAQ
Milestones and honest status
| Milestone | Scope | Status |
|---|---|---|
| M0 | offline registry, questions, acceptance drafts, revisions, evidence, reports, loopback API | implemented, tested on Windows/Node 24 (see test report) |
| M0.2 | independent audit with sealed JSON/Markdown/HTML reports, verify, CI smoke test, proof pack | implemented in 0.2.1 |
| M1 | opt-in isolated runner that captures command, commit SHA, environment, exit code, logs and hashes; richer importer | planned |
| M2 | authenticated review (OIDC), tamper-resistant approvals | planned |
| M3 | Git/CI change graph and stale-impact analysis | planned |
| M4 | accessible dashboard with usability testing | planned |
| M5 | PostgreSQL, tenant isolation, encrypted evidence, external signed audit | planned |
| M6 | verified connectors (GitHub, ClickUp, Jira, OpenSpec, Spec Kit), per-IDE activation tests | planned |
Do not read M0 passing as "product released at enterprise grade".
FAQ
Does it run my tests? No. You import the outcome, and it is labelled not independently verified. A runner is M1.
Why can I never mark a requirement "complete"? Because a tool cannot know that. It reports what is observed, imported, attested, stale or missing, and a human decides.
Is the audit a certification? No. It is an evidence-integrity review of your own records.
Can someone fake an approval? With local write access, yes, by editing files. The audit catches silent edits to the registry (AUD-08) and the log (AUD-01) unless the whole record is rewritten. For stronger assurance, keep the folder in Git with protected branches and review.
Does it need an AI key? No. Skills and agents are optional instruction files for assistants you already use.
Do the agents run automatically? They are portable instruction files. Native activation in each IDE is not verified.
Windows support? Yes; tested on Windows. File-symlink tests are skipped there unless Developer Mode is on.
Which Node versions? 22 and 24 in CI.
How do I start with a vague idea? ingest a short Markdown note; read the questions; answer them in a revised file; approve.
Where is my data? .intentsutra/ in your workspace; nothing is uploaded.
How is this different from a story-writing AI? It produces fewer words and more accountability: stable IDs, version-bound evidence, human gates and an audit trail.
Which license? Apache-2.0.