optimize audit sources → ledger → AEM autofix

optimize does not invent its own audit. It aggregates existing audit skills into one findings ledger, then autofixes the platform-fixable findings for AEM. This file is the mapping: each source, how to run it, how its findings normalize into the ledger (layer + fixability), and which AEM fixer (if any) resolves them.

All sources share one ledger (optimize/findings.json), one id space (f-<hash(source|layer|check|scope)>), one scorecard, one gate, and one loop. A run of a source only resolves its own findings — never another source's.

The sources

These are referenced, not vendored (the user's decision). impeccable is already a stardust dependency; the marketing skills come from coreyhaines31/marketingskills. If a source isn't installed, skip it and note the gap — the baseline source always runs.

Recording an external finding

The agent runs a source, then normalizes each of its findings:

node skills/rollout/scripts/findings.mjs record \
  --source <source> --layer <layer> --check <check-id> \
  --severity P1|P2|P3 --fixability platform-migration|design-pass|out-of-scope \
  --scope-level page|site --scope-ids <slug[,slug]> \
  --evidence "what was observed" --recommend "the fix"

Normalization rules of thumb:

• Severity: the source's "critical / high" → P1, "medium" → P2, "low" → P3.
• Fixability: can the EDS delivery fix it (head/metadata/blocks/CWV) → platform-migration; is it authored content/design → design-pass; infra/external → out-of-scope.
• check id: reuse a stable, descriptive slug; if it matches a row in the AEM fixer table below, autofix will pick it up automatically.

Fixability → who fixes it

• platform-migration → the AEM autofix engine and/or a re-deploy resolve it.
• design-pass → upstream (migrate / prototype); rollout surfaces only.
• out-of-scope → informational; never gates beyond its severity.

AEM autofix registry (autofix-aem.mjs)

check → fixer. kind: deterministic (mechanical) · content-draft (generates copy, applied under the aggressive policy, logged for review) · manual (prepares guidance/payload, a human applies).

A finding whose check isn't in this table gets autofix.available = false (surface only). New fixers are added by extending AEM_AUTOFIX in lib.mjs and adding the fixer body in autofix-aem.mjs — no schema change.

The loop

run sources → record into ledger → optimize gate (open P1?)
        ▲                                   │ open P1
        │                                   ▼
   re-verify ◄── re-deploy ◄── autofix-aem (edits EDS project, stages in-progress)

autofix stages applied findings in-progress; a re-deploy + re-optimize (or re-record for external sources) confirms the fix and flips them to fixed. The gate is clean when no open P1 remains from any source.