Skill 6 of 6 · Orient
deep-docs
/deep-docs
A standing, queryable map of what your codebase has actually built - priced for agents on a token budget, verified claim by claim.
What it does for you
Docs your agents can afford to read
Runs after a build or standalone on any repo. It documents what exists - it never edits source.
Load only what you need
Three tiers: a tiny always-loaded index, per-subsystem cards loaded on touch, deep references on demand. An agent pays only for the slice its task touches.
Anchored or it's fiction
Every claim cites a real file, line, and symbol in your codebase. No intent-prose that describes what the code was supposed to do instead of what it does.
Verified before publish
Fresh agents re-check every single anchor against the code - no sampling. One dead anchor blocks its card from publishing. If the map says it, the code shows it.
Stays current cheaply
Refresh detects drifted anchors and undocumented additions, then regenerates only the stale and new subsystems - not the whole map.
The guarantee
Every anchor checked - no sampling
Anchor verification is exhaustive because the guarantee is binary: if even one claim is unverified, the whole map can't be trusted. So every claim gets checked, every time.
Honest about its edges. Alongside the map, a coverage report states exactly what was documented, what wasn't and why, and which existing human docs were referenced but never absorbed as verified fact.
Where it fits
The last stop - and the head start on the next
Once the code is verified, deep-docs maps what got built - so the next effort starts oriented, not lost.