From 63549f73eeedb197c856647f378912521445931f Mon Sep 17 00:00:00 2001 From: Henry Lach Date: Wed, 16 Sep 2026 01:17:44 -0400 Subject: [PATCH] =?UTF-8?q?docs(skill):=20create-taskplane-task=201.3.0=20?= =?UTF-8?q?=E2=80=94=20the=20worker=20self-check=20reads=20Completion=20Cr?= =?UTF-8?q?iteria;=20never=20pre-seed=20Self-check=20sections=20(#657)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- skills/create-taskplane-task/SKILL.md | 22 ++++++++++++++++++- .../references/prompt-template.md | 5 +++++ 2 files changed, 26 insertions(+), 1 deletion(-) diff --git a/skills/create-taskplane-task/SKILL.md b/skills/create-taskplane-task/SKILL.md index 7c11068a..24bb9f64 100644 --- a/skills/create-taskplane-task/SKILL.md +++ b/skills/create-taskplane-task/SKILL.md @@ -1,6 +1,6 @@ --- name: create-taskplane-task -version: 1.2.0 +version: 1.3.0 description: Creates structured Taskplane task packets (PROMPT.md, STATUS.md) for autonomous agent execution via the task-orchestrator extension (/orch). Use when asked to "create a task", "create a taskplane task", "stage a task", "prepare a task for execution", "write a PROMPT.md", "set up work for the agent", "queue a task", or whenever the user wants to define work that will be executed autonomously by another agent instance. --- @@ -486,6 +486,26 @@ pipeline. - Include at least one NEW test file per task — workers can't shortcut test creation - Each step's artifacts list must include at least one source file (not just STATUS.md) +### The worker's self-check reads what you write here + +Before every code review the worker must write a `### Self-check (Step N)` +table into STATUS.md (`review_step` refuses without it) with one row per +step outcome, per Step 0 design decision the step implements, and per +`## Completion Criteria` item the step touches — each with `file:line` +evidence. Two consequences for task creators: + +- **Do NOT pre-seed `### Self-check` sections in STATUS.md.** The runtime + requires the section to be written *after* the step's last checked box and + to contain evidence rows; an empty or early-placed section is refused. +- **Make `## Completion Criteria` concrete.** The template's defaults ("all + steps complete / tests passing / docs updated") give the self-check nothing + to verify. Add the task-specific, checkable outcomes ("`review_step` refuses + when the section is missing", "resume refuses a `.DONE` over a REVISE") — + these become rows the worker must produce evidence for, and the reviewer + verifies that evidence instead of re-deriving the review. For Review Level + ≥ 2 tasks, a Step 0/1 that records design decisions in a STATUS table pays + for itself the same way. + --- ## Key Principles diff --git a/skills/create-taskplane-task/references/prompt-template.md b/skills/create-taskplane-task/references/prompt-template.md index 362188d3..7e9f988e 100644 --- a/skills/create-taskplane-task/references/prompt-template.md +++ b/skills/create-taskplane-task/references/prompt-template.md @@ -145,6 +145,11 @@ Review Level 0 is ONLY for trivial changes. Most M+ tasks need Level ≥1. ## Completion Criteria +> The worker's pre-review self-check enumerates these. Replace the generic +> lines with task-specific, checkable outcomes the worker can cite evidence for. + +- [ ] [Task-specific outcome, e.g. "`review_step` refuses when X is missing"] +- [ ] [Task-specific outcome] - [ ] All steps complete - [ ] All tests passing - [ ] Documentation updated