check

ds:check — drift detector

Read-only. Validates DOSSIER.md against reality. Returns severity-tagged violation list. Never mutates.

Steps

0. Detect host env

Per ADAPTERS.md. Note HAS_CTX (parallel scout dispatch benefits).

1. Locate

Live dossier per ds:status step 1. If --all: also check archived dossiers (sanity sweep).

2. Plan scout dispatch

For each repo in §X: prepare a scout mission. Missions are independent + parallel.

Mission template (per scout):

You are a dossier-scout. Read-only.

Dossier: <path-to-DOSSIER.md> (full file pasted below)
Repo to scan: <repo-path>

Tasks:
1. For each §V row with `check` = shell predicate: run the predicate, report PASS/FAIL.
2. For each §V row with `check` = test name: locate test in <repo-path>, run if executable, report PASS/FAIL/MISSING.
3. For each §T row with state=`x` + `cite=<sha>`: verify <sha> exists in <repo-path> git log. Report VALID/MISSING.
4. Refresh §X for <repo>: git status -sb, ahead-count, latest tag, push state. Report current values; do NOT modify DOSSIER.md.

Output: caveman pipe-table. One row per finding.

<paste DOSSIER.md here>

Spawn one dossier-scout per repo, in parallel. Use Agent tool with subagent_type: dossier-scout.

3. Local checks (main thread)

Independent of scouts. Run concurrently with dispatch:

Vm	Check	How
Vm.1	live dossier count ≤ 1 per slug	grep INDEX for duplicate slugs
Vm.2	every §S line has ISO timestamp	regex ^[0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}
Vm.3	every §T x row has non-empty cite	awk §T table
Vm.4	closed dossiers under _archive/	ls live dirs for §Z presence
Vm.5	INDEX counts match DOSSIER §T/§B	parse both, diff
Vm.6	no §S START without DONE for same target	awk pairing
Vm.7	INDEX regenerable (run lib-regen-index.sh, diff against current)	optional
Vm.8	no .tmp orphans	find .scratchpad/dossier/**/*.tmp
Vm.9	locks not stale	run lib-clear-stale-locks.sh in dry mode
Vm.10	migrate markers consistent (if migration in progress)	n/a unless ds:migrate active

4. Aggregate

Collect:

• Scout reports (one per repo).
• Local Vm findings.

Group by severity:

Severity	Examples
🔴 critical	§V failing + commit cited (spec lies); §T x-row commit missing (history rewritten)
🟡 warning	§X stale (>1 day); Vm.6 incomplete op; §V test MISSING (untestable claim)
🟢 info	§X mtime close to current (refresh anyway); INDEX count off by 1

5. Report

ds:check <slug>: <N> critical, <M> warnings, <K> info

🔴 critical
  <repo>: <finding>
  <repo>: <finding>

🟡 warnings
  ...

🟢 info
  ...

Vm summary:
  Vm.1 ✓ Vm.2 ✓ Vm.3 ✗ (2 rows) Vm.4 ✓ ...

Suggested remediations (do NOT auto-apply):
  - <T-id> cite=<sha> missing → ds:backprop or rebase
  - §X <repo> stale → ds:build --next (refresh as side-effect)
  - §V.<N> test MISSING → ds:backprop B<N>

6. No mutations

ds:check does NOT write to DOSSIER.md, INDEX.md, or any repo. Period.

Append §S? No. Read-only verb. Skipping §S keeps the log noise-free.

Exception: if lib-regen-index.sh is run as part of Vm.7 dry-check, that's a derived-state regen (idempotent, not a mutation per Vm.7).

Failure handling

• Scout timeout / refuse: report scout-failed for that repo. Other scouts still proceed.
• Repo missing: flag in report, do not error.
• DOSSIER.md missing: refuse w/ "no live dossier".

Cite

• FORMAT.md §17 (Vm rules)
• ADAPTERS.md §context-mode
• agents/dossier-scout.md (subagent contract)