20260615 Stale Codex Guidance Cleanup

Goal

Patch stale Codex orchestration guidance blockers in the lazycodex-gate-reviewers worktree.

Skills And Tier

• programming: used for the TypeScript/MJS edits and test updates.
• LIGHT: scoped wording/marker cleanup inside existing guidance, sync script, and direct tests; no new runtime boundary, module, abstraction, auth, DB, or concurrency change.

Success Criteria

1. Shared start-work guidance uses current Codex v1 multi_agent_v1.spawn_agent examples with fork_context:false and current reviewer role naming.
2. Ulw-loop aggregate goal instructions say not to call update_goal mid-aggregate, checkpoint the OMO ledger story, and reserve update_goal for the final story after the quality gate.
3. Legacy compatibility cleanup still removes old generated guidance while active source/test-support no longer trips stale literal scans.

Evidence

• RED: .omo/evidence/20260615-stale-codex-guidance/red-codex-goal-instruction-test.txt captured the updated direct ulw-loop test failing against the stale wording.
• GREEN: .omo/evidence/20260615-stale-codex-guidance/green-codex-goal-instruction-test.txt shows 12/12 direct ulw-loop instruction tests passing.
• GREEN: .omo/evidence/20260615-stale-codex-guidance/node-sync-skills-tests.txt shows 19/19 sync-skills tests passing.
• GREEN: .omo/evidence/20260615-stale-codex-guidance/ulw-loop-typecheck.txt shows tsc --noEmit completed.
• GREEN: .omo/evidence/20260615-stale-codex-guidance/plugin-check.txt shows bun run --cwd packages/omo-codex/plugin check passed with 217/217 plugin tests.
• GREEN: .omo/evidence/20260615-stale-codex-guidance/test-codex.txt shows bun run test:codex passed with 339/339 Codex tests and 95/95 lsp-tools tests.
• SCAN: .omo/evidence/20260615-stale-codex-guidance/final-stale-scan.txt records no matches for the requested stale-token scan.
• SCAN: .omo/evidence/20260615-stale-codex-guidance/extra-targeted-stale-scan.txt records no matches for the broader targeted stale-token scan.
• SCAN: .omo/evidence/20260615-stale-codex-guidance/evidence-path-redaction-scan.txt shows committed evidence was checked for local home/worktree path leaks after redaction.

Self Review

Re-read the diff after verification. The active generated guidance no longer mentions the v2 fork-mode wording, stale compatibility sections are removed by the Markdown section boundary rather than hidden literal-token construction, and tests assert behavior rather than reconstructed implementation tokens. The plugin package check and full Codex compatibility gate both passed.