Seek parallel board work and save only missing recovery state

[EHotwagner]

Require work-board to actively seek useful parallel work at entry/resume and material transitions, checking actual prerequisites, owners/PRs, touch-sets, shared outputs and remaining resources. Preserve serial expensive-run limits, one owner per attempt and one integrator per shared surface. Keep the explicit unfiltered intake triage already delivered by PR4700.

Stop/save uses existing issue/native evidence and Atlas pointers when sufficient; a separate handoff and save PR are needed only for missing recovery state. Align continuation and skill mirrors/UI metadata, retain shutdown/custody/admission safeguards and capture the decisions in Atlas. This is skill maintenance, not a request to stop or dispatch programme work.

Validation: skill frontmatter, local links and mirror equality; registry preparation; Atlas archive/history validation; final native CI. No publication or installed fleet activation is claimed. A pre-existing product-inspection example is completed with its required binding/report flags, retaining the original semantic-quality failure39271932096.

[EHotwagner]

Independent exact-head review: PASS for a499851248734bb0db874d15b28dc25d2439b008, tree ff69f77bd7d2f86b3ba1832c9f51004919870afa.

Reviewer /root/review_creator_delta was separate from implementation and performed read-only source review. Parallel-opportunity checks remain bounded by actual capacity, ownership, isolation and admission. Stop/resume retains terminal shutdown, original uncertain effects and durable recovery pointers. The #4700 intake triage remains intact. Mirrors, metadata, Atlas citations and manifest hashes agree; no findings, suite reruns or remote mutations by the reviewer.

Normal delivery remains responsible for fresh exact-head checks and merge readback. No publication is selected. p-add is unchanged and its legacy organization instructions remain a V2 migration follow-up; pnext-item remains the single-item execution route.