task-checker
Task Checker
Goal
Turn “is this plan task actually done?” into a reviewable evidence chain:
- Per-task verification: confirm goals/acceptance criteria/deliverables exist in code and observable behavior
- Per-task drift detection: confirm the task description still matches the implementation (stale / diverged / re-scoped)
- Per-task test review: assess whether tests cover key success paths and failure/edge cases; propose minimal additions
Hard Rules (No False Positives)
These rules exist to prevent “looks done” audits that miss real gaps.
1) No fabricated or “hand-wavy” evidence
Disallowed evidence patterns (treat as a GAP):
- “Implicitly tested via integration” without a concrete test file + assertion that exercises the exact acceptance point
More from satone7/skills
writing-plans-plus
Enhanced planning with structured, machine-readable task format based on writing-plans. Use this skill whenever you need to create implementation plans with explicit completion tracking, cross-validation issue tracking, or machine-readable JSON output. Always use this when planning software development tasks, especially when tasks need to be executed by other agents or reviewed later.
10find-next-task
Find the next executable task from writing-plans-plus JSON plan files. Use this skill whenever you need to locate the next task to work on, check task status, or understand what to do next in a project with structured plans. Always use this when the user asks "what's next", "find the next task", "show pending tasks", or wants to continue working on a plan.
5aitc-workflow
>
5guardian
|
4find-task-skills
>
4executing-single-task
执行 writing-plans-plus 计划中的“恰好一个任务”:输入为 find-next-task 的完整输出 JSON。严格按四阶段流程(实现→规范审查→质量审查→最终验证),基于证据更新 plan JSON 的 passes(成功时删除 issue 字段;失败时写入/更新非空 issue[]),并创建且仅创建一个 git commit。凡用户要求“执行下一个任务/只跑一个任务/把 passes 写回/自动执行单任务”都应优先使用本技能。
4