Request: Skill Small Improvements Batch (2026-05-29)

Request Summary

Apply a small, low-risk batch of skill/tooling improvements that reduces repeated command construction and prompt duplication without changing product runtime behavior.

Why This Batch Exists (Evidence)

• The tracker shows frequent repeated focused-test command construction with long dotnet test --filter strings and repeated validator triads (Scripts/Test-TaskRecords.ps1, Scripts/Test-PageLinks.ps1, Scripts/Test-PagePathLiterals.ps1) in many rounds.
• CI monitoring guidance is currently duplicated across .github/agents/smith.agent.md, .github/skills/ci-status-monitor/SKILL.md, and .github/skills/training-sprint-loop/SKILL.md.
• Skill inheritance and MCP-first adoption now exist, but there is no lightweight repo check that enforces those contracts.

Proposed Small Changes

1. Add a focused test command builder hook

• Proposal: add Scripts/SkillHooks/New-FocusedDotnetTestCommand.ps1.
• Purpose: generate a stable dotnet test ... --filter command from test names and optional categories.
• Output contract: machine-readable JSON with command, filter, and warnings.
• Classification: Now.
• Impact: medium (fewer command typos and shorter chat tokens).
• Effort: low.
• Confidence: 93%.

2. Add a compact "skill contract" verifier

• Proposal: add Scripts/Test-SkillContracts.ps1.
• Purpose: verify each implementation-focused skill includes Inherits from task-core-loop, and each user-invocable skill includes an output section.
• Scope: structural/lint checks only, no behavior checks.
• Classification: Now.
• Impact: medium-high (prevents contract drift).
• Effort: low-medium.
• Confidence: 90%.

3. Collapse duplicated CI-mode wording into one reference block

• Proposal: keep authoritative CI-mode policy in .github/skills/ci-status-monitor/SKILL.md and shorten repeated copies in other skills to one-line references.
• Classification: Next.
• Impact: medium (token and maintenance reduction).
• Effort: low.
• Confidence: 88%.

4. Add a tiny schema README for hook outputs

• Proposal: add Scripts/SkillHooks/README.md documenting stable output keys for Get-CiSnapshot.ps1, Wait-PrReviewState.ps1, New-TaskEvidenceComment.ps1, and New-CouncilEvidenceBundle.ps1.
• Classification: Later.
• Impact: medium (easier reuse by future skills/agents).
• Effort: low.
• Confidence: 85%.

5. Add a model-profile bootstrap hook for trained/default + stock comparator setup

• Proposal: add Scripts/SkillHooks/Ensure-TrainingModelProfiles.ps1.
• Purpose: verify required Ollama tags exist, optionally create a stock 24k comparator tag, and write a stable pair of model profiles (trained default + stock comparator) into the resolved settings override.
• Output contract: machine-readable JSON with resolvedSettingsPath, ollamaTags, createdTags, profileIds, and warnings.
• Classification: Now.
• Impact: high (removes repeated manual setup drift before A/B sessions).
• Effort: medium.
• Confidence: 92%.

Status

InProgress

Execution tracking:

• TSK-0251 - focused dotnet test command hook
• TSK-0252 - skill contract lint script
• TSK-0253 - training model profile bootstrap hook

Follow-Up

• Now:
• implement change 1 via TSK-0251 (confidence 93%)
• implement change 2 via TSK-0252 (confidence 90%)
• Next:
• collapse duplicated CI wording to shared references (confidence 88%)
• Later:
• add stable hook output schema README and examples (confidence 85%)