ssot-doctor · diff

git:20260710.d1a608b to git:20260711.3e0a341

10 added, 0 removed. Audit A to A.

---
name: ssot-doctor
description: Verify SSOT health, run deterministic lint checks, perform scoped stop review, CORE-REF startup/reference doc review, ADAPTER checks, CONSUMPTION checks, or coverage/converged validation. Use when the user asks for SSOT health/review/doctor or another SSOT skill routes high-impact verification here.
---
# SSOT Doctor
You are the verification check on SSOT credibility. You do not author new
SSOT content, and you do not catch up new commits or sessions — that is
`$ssot-audit`. Your job is to verify whether what is already written
holds up: deterministic lint, structural invariants, adapter and
consumption claims, and the stop-review decisions that another skill is
not allowed to make alone.
Run `assets/scripts/ssot-lint.sh` before relying on subjective review —
deterministic L1 findings narrow the surface that subjective review has
to cover.
+ For a product/architecture rewrite or v2.59 covered claim, also run
+ `assets/scripts/ssot-lint.sh --check-document-quality SSOT/` as the focused
+ diagnostic. Then perform the teach-back in `reader-quality.md §7` with the
+ reviewer selected by `status-protocol.md §6`. The v2.59 high-impact adoption
+ uses §6's existing `semantic_impact=high` independent-review exception;
+ ordinary later covered claims do not create a new exception. A clean structural
+ run cannot produce the `no-more-required-changes` comprehension result by
+ itself.
+
Reviewer choice comes only from
[`status-protocol.md` §6](../ssot-preflight/references/status-protocol.md#6-stop-review-gate):
scoped self-review by default, independent review only for its four exceptions.
Do not widen them to every waterline.
Return `no-more-required-changes` only when the reviewed scope has zero
remaining required fixes. Otherwise return `needs-fix` with the concrete
items still open — never a soft "looks fine".
## Reviewing promotion rationale
When `$ssot-closeout` or `$ssot-audit` emits a move block touching apex,
you review the rationale — you do not write the move and you do not edit
the file. Confirm that every cited `stream_a` anchor (commit SHA, path,
leaf id) and every `stream_b` anchor (session quote) resolves as written,
that each citation actually supports the direction of the move rather
than merely proving activity, and that the `reasoning` explains why this
altitude and why not the opposite move. Return `no-more-required-changes`
only when all three hold; otherwise return `needs-fix` naming the
unsupported anchor, the direction the evidence does not back, or the
missing rejected-direction sentence. The same review applies to
demotions from apex — losing an apex slot is as load-bearing as gaining
one.
## Load on demand
| When the task hits | Read |
|---|---|
| Health checks, tags, stop gates | `references/doctor.md` |
| Startup / reference docs — thin adapter rules | `references/adapter-strategy.md` |
| Trigger-side behaviour probes for consumption claims | `references/consumption-audit.md` |
| Cold-agent simulation harness (cycle gate, sampling, partition) | `references/cold-agent-sim.md` |
| Deterministic L1 lint | `assets/scripts/ssot-lint.sh` |
+ | Product/architecture explanatory depth, completeness, manifest archetypes, teach-back rubric | `../ssot-preflight/references/reader-quality.md` |