| name | task-core-loop |
| description | Shared core workflow for MemorySmith task execution: scope, evidence, minimal-change implementation, validation, and tracking updates. |
| argument-hint | Task goal, scope boundaries, and validation target |
| user-invocable | false |
| disable-model-invocation | false |
Task Core Loop
Use this internal skill as the shared base for implementation-focused workflows.
Core Steps
- Confirm scope and non-goals.
- Gather source-of-truth evidence from code, wiki memories, and active docs (MCP tools first).
- Implement the smallest reversible slice.
- Run the narrowest validation that can fail.
- Record evidence, assumptions, confidence, and open questions.
- Update tracker and any durable memory notes affected by new facts.
Core Guardrails
- Keep changes minimal and focused to the stated slice.
- Prefer
/tasks traceability for implementation work.
- Prefer MCP tools over manual file rewrites when tool coverage exists.
- Avoid repetitive polling loops when script-driven waiting or snapshot checks are available.
- Default to in-process analysis; invoke subagents only with explicit user permission in the current request.
Modular Hooks
- CI snapshot:
pwsh ./Scripts/SkillHooks/Get-CiSnapshot.ps1 -Branch <branch> -Commit <sha>
- PR bounded wait:
pwsh ./Scripts/SkillHooks/Wait-PrReviewState.ps1 -PullNumber <number> -PollSeconds 60 -TimeoutMinutes 20
- Task evidence comment payload:
pwsh ./Scripts/SkillHooks/New-TaskEvidenceComment.ps1 ...
- Council evidence bundle payload:
pwsh ./Scripts/SkillHooks/New-CouncilEvidenceBundle.ps1 ...
Core Outputs
- Scope summary.
- Evidence list.
- Validation result.
- Residual risks and next step.
