Restore Emil critique layer from the CMO original at full fidelity

[TheGreatAxios]

Restores Emil's CMO prompt layer at full fidelity: ports back every section the design-eng overhaul narrowed away (capabilities, Thinking and Reasoning laws, Boy Scout Rule, cross-reference checklist, temp-test workflow, full report format, guidelines, negative constraints). Tests assert the restored contract; prompt-size budgets raised deliberately.

Source pin: abklabs/agents plugins/cmo/agents/emil.md @ e1d626cc6b7c6c2911a94daccd7793cc03633e26 (2026-04-21; only commit ever touching that file).

Deviations from the source (deliberate, exhaustive; also in the package header):

1. Fleet framing (EmilDirector wrapper + PRIMARY INTENT), same job.
2. Lane routing to builder/draper/rand/critic/greybeard; source knows no fleet.
3. BLINDERS ON brief-scoping kept from the overhaul.
4. skills: brand-identity has no Corbits equivalent; substance inlined.
5. Report format gains Blockers/Paths per the worker envelope.
6. Temp-test path tmp/critique-tests/ kept verbatim.
7. Do-not-commit kept verbatim; commits stay parent-owned.
   Fleet fields unchanged (deliberate): maySpawn false, REVIEW_TOOLS, modelRole review, tier leaf; source model: sonnet is an agents-repo pin, not carried over.

Terminology note: the ticket says 'copy/messaging layer'; the pinned source contains no copy/messaging content (it is the design-eng critique agent housed in the CMO plugin), so this restores the prompt copy at full fidelity and adds no marketing lane. Flagging in case the ticket meant something else.

Verification: bun test src/agent/directors/emil/ 20 pass; bun run typecheck exit 0; full suite bun test ./src ./tests ./evals ./scripts --randomize --seed 424242 7139 pass / 0 fail; bun run check exit 0.

Fixes CL-7800

[linear-code]

CL-7801

CL-7800

[TheGreatAxios]

Audit: the CMO-original restore content (capabilities, Thinking & Reasoning laws, Boy Scout Rule, cross-reference checklist, temp-test workflow, guidelines, negative constraints) is faithful and worth landing. One blocking issue:

Envelope ownership: #968 establishes that the scaffold owns the worker report envelope (Summary/Findings/Blockers/Paths) and lane packages must not re-specify it — and #968 wins that conflict. The restored # Report format section in emil/package.ts re-specifies an 8-heading in-package format (## Summary / ## Findings / ## Test results / ## Recommended tests / ## Blockers / ## Observations / ## Paths plus Confidence/Severity), which directly contradicts #968. Please strip it down to the scaffold envelope and carry the lane extras (test results, recommended-tests, observations, confidence/severity) as Findings sub-bullets, the way the other lane packages do. The package.test.ts "restores the full report format" test needs the same treatment.

[TheGreatAxios]

Audit (fleet builder): envelope ownership — #968 WINS, yield here.

The scaffold owns the worker report envelope (Summary / Findings / Blockers / Paths), so in-package prompt formats must not re-specify headings. This PR's # Report format 8-heading section (## Test results, ## Recommended tests, ## Observations, etc.) collides with the scaffold; fix pushed in this branch: # Report now defers to the scaffold envelope and carries the restored CMO report content as Findings sub-bullets (per-issue law/location/evidence/confidence/severity, test results, recommended tests, observations). Deviation comment #5 updated to say the format yields to the scaffold (was: gains Blockers/Paths). Test rewritten to assert no re-specified headings + content preserved. No new envelope headings invented — lane extras live as Findings sub-bullets per the shared rule.

[TheGreatAxios]

Superseded by #999 (fresh rebase under CL-7800's name, release 0.3.26).