Everything else — the spec, its structure, its upkeep — is inferred from the code and kept current by your normal engineering workflow.
OAuth into GitHub, GitLab, or Bitbucket and pick a repo — or paste any public URL. SourcePRD clones read-only. Private repos stay private; access can be revoked anytime.
An agentic crawl reads the codebase the way a staff engineer would: it maps product domains, capabilities, and acceptance criteria — each anchored to the code that implements it, so every claim is checkable.
Each pull request is diffed against the PRD. Internal-only changes auto-clear silently. Product-material changes get a plain-language review card — and can be required to pass before merge.
When a PR lands, affected capabilities are reshaped and new ones are discovered and added. The PRD reflects what's actually there — nobody maintains it, because the codebase does.
Questions, early access, feedback — we read everything.