feat(update,fix,doctor): auto-populate markers, validate YAML/model ids, add agentrig fix + --auto-fix

[doidor]

What this PR does

Implements all 9 improvements reported from the @doidor/markbook agentrig update experience, so the next update on any target repo no longer needs the surgical hand-patching that motivated the report.

#	Item	Implementation
1	Marker populator — the <!-- AGENTRIG:skills-inventory:start --> block in AGENTS.md is auto-rewritten by both compile and update; enumerates .agents/skills/ on disk so user-added skills (e.g. the 6 markbook-* skills in this very repo) appear alongside AgentRig-bundled ones	src/core/markers.ts (new), src/core/install.ts (filesystem walker)
2	YAML validation gates update — uses parseAllDocuments so multi-doc scenario files are tolerated; broken state-machine.yml now blocks the update	src/core/validate.ts (new)
3	Model-ID validation against the SDK registry — live listModels() when available, curated allowlist as offline fallback, "did you mean…" suggestions	src/core/validate.ts, surfaced in doctor and update
4	agentrig fix + agentrig update --auto-fix — deterministic repair engine; restores broken YAML from canonical and replaces unknown model ids (e.g. the retired gpt-5) with the safe auto fallback; writes .bak backups; no agent / network needed	src/core/fix.ts (new), src/commands/fix.ts (new)
5	--diff drift classification — preserved files tagged 🔴 broken / 🟡 stale / 🟢 enhancement / ⚪ mixed with a per-severity summary at the top	src/commands/update.ts:diffPreserved
6	--skip-agent enumerates preserved files inline — no more "11 files preserved" without naming them	src/commands/update.ts:logFileList
7	New files grouped by parent dir in the post-install summary	same
8	doctor detects npm-link installs + checks npm latest — linked-checkout vs registry provenance, surfaces validation findings (YAML / model-id) as exit-blocking failures	src/commands/doctor.ts:detectInstallProvenance
9	Reconciliation history — state.json.reconciled[] records per-file decisions + the canonical hash at decision time; the next update skips re-prompting on files the user consciously diverged on, unless canonical has drifted past the recorded hash	src/core/state.ts:ReconciledRecord, src/commands/update.ts:alreadyDivergedOn

Audit upgrade

• New marker-populated check type in both src/core/audit.ts and knowledge/templates/eval/static-audit.mjs so CLI + in-repo audit stay in lock-step.
• agents-skills-inventory upgraded from a loose file-contains substring match to marker-populated — the audit now fails (correctly) if the block is empty, has unfilled placeholders, or omits any skill under .agents/skills/. This is what surfaced the disk-enumeration bug fixed in Automate releases via Changesets #1.

Bonus find logged as a gotcha

The new strict check immediately caught a latent bug: the original skillsInventory() only iterated manifest.artifacts, so user-added skills under .agents/skills/ were silently dropped from the populated block. Fixed and logged in .agents/wiki/skills-inventory-populator-enumerates-disk.md per critical-rule #2.

Docs

• README.md — new fix row + --auto-fix flag in Commands table; new "Self-healing the install" subsection.
• docs/commands.md — extended update, compile, doctor; new fix section.
• docs/getting-started.md — iterate block now includes --auto-fix + fix.
• knowledge/templates/AGENTS.md — notes that the skills-inventory marker is auto-populated.

Verification

• npm run build clean under TypeScript strict.
• CLI audit + in-repo static-audit.mjs both 100% / 100%.
• Round-trip:
  1. Broke .agentrig/harness/state-machine.yml (appended invalid YAML).
  2. Injected model: gpt-5 (retired ID) into .agentrig/agents/developer.yml.
  3. eval --static → P1 + P2 FAIL.
  4. doctor → surfaced both validation findings + "did you mean…" suggestions + linked-checkout detection.
  5. agentrig fix → restored state-machine.yml from canonical, replaced gpt-5 with auto.
  6. eval --static → back to 100%.
• Knowledge bundle bumped 0.5.0 → 0.6.0; Changeset at .changeset/update-workflow-improvements.md (minor bump).

Out of scope

• New principles or rules.
• Dynamic eval pipeline.
• Provider abstraction beyond the existing validateModel hook.