菜单
Run a /tasks-first implementation loop from active task selection through validation evidence and task status/comment updates.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Monitor CI conservatively with snapshot-first status checks and actionable failure triage.
Submit a pull request, request review, triage comments, fix issues, and use script-driven review waits with a bounded timeout before merge/cleanup. Use when implementation is complete and the branch is ready for PR handling end to end.
Audit and reconcile drift between agent/prompt docs and actual runtime capabilities in MemorySmith services/controllers.
Run a repeatable hygiene audit over Data/Memories and Data/Pages for drift, stale facts, broken source links, and cleanup candidates.
Run a MemorySmith LLM council review for high-impact decisions. Use when evaluating schema changes, retrieval/search behavior, chat or agent write governance, wiki conventions, or long-term architecture tradeoffs with explicit dissent, risks, confidence, and acceptance criteria.
Run an agent self-review for skill and prompt improvements, including script extraction opportunities and token-conscious workflow upgrades.
| name | task-delivery-sprint-loop |
| description | Run a /tasks-first implementation loop from active task selection through validation evidence and task status/comment updates. |
| argument-hint | Task ID/key, change slice, and validation scope |
| user-invocable | true |
| disable-model-invocation | false |
Inherits from task-core-loop.
Use this when delivery should be anchored to /tasks records, not ad hoc notes.
Use MCP task tools first (task_get, task_set_status, task_add_comment) and use scripts to generate structured evidence payloads instead of writing long comments by hand.
InProgress when implementation starts via MCP status tool.task-core-loop.pwsh ./Scripts/SkillHooks/New-TaskEvidenceComment.ps1 -TaskId <id> -Summary <summary> ...