Skill and Agent Prompt Requests Master List (2026-05-29)

Status legend: Requested | InProgress | Implemented | Deferred

Outstanding and Recent Requests

1. Core task inheritance across skills - Implemented
2. Token-conscious PR review hooks and script waits - Implemented
3. Self-review skill and governance loop - Implemented
4. MCP-first modular extraction batch 2 - Implemented
5. MCP tool surface recommendations - Requested
6. MCP authoring usability and chat tool parity - Requested
7. Skill small improvements batch - Requested
8. PR review closure and thread sync skill - Deferred (covered by canonical pr-review-delivery + hook-script follow-ups)
9. Script extraction backlog for additional repeated terminal flows - Requested
10. Prompt/runtime parity audit cadence standardization - Requested
11. Wiki-memory hygiene audit cadence standardization - Requested

2026-05-30 Self-Review Follow-Up

1. Skill small improvements batch - InProgress
2. MCP authoring usability and chat tool parity - InProgress
3. Focused dotnet test command hook execution task (TSK-0251) - Requested
4. Skill contract lint script execution task (TSK-0252) - Requested
5. Model profile and Ollama comparator parity for training A/B - InProgress
6. Model-profile bootstrap hook for trained/default and stock comparator (TSK-0253) - Requested
7. Canonicalize PR review skill in repo-local skills and keep repo as skill source of truth - Implemented

Now / Next / Later Snapshot

• Now (high confidence): script hooks for repeated focused test command generation (TSK-0251, confidence 93%) and skill contract linting (TSK-0252, confidence 90%).
• Now (high confidence): add model-profile/Ollama bootstrap automation for trained default plus stock comparator parity (TSK-0253, confidence 92%).
• Next (medium-high confidence): parity drift automation and coding-agent surface synchronization under TSK-0224 (confidence 90%).
• Later (medium confidence): cadence refinements for prompt/runtime and hygiene audits once script hooks are stabilized (confidence 85%).

2026-05-31 Skill Pack Rationalization Snapshot

• Skill count target adopted: keep repo-local skills within an 8-10 range.
• Canonical source policy adopted: PR review delivery skill now lives in .github/skills/pr-review-delivery/SKILL.md.
• Consolidation applied: retired narrow standalone skills gpu-reality-validation and training-contract-evolution in favor of broader training loop coverage.

Maintenance Rules

• Add all new skill/agent-prompt change requests here first.
• Use a dedicated request page when scope is significant.
• Group small related changes into one request page when possible to avoid page clutter.
• Mark status updates in both this master list and the linked request page.