00.Feat 260525015.Done.Qa Subagent And Validation

---
id: '260526006'
type: task
status: done
priority: '00'
slug: create_validate_task_skill
title: Wire /agn:validate task lightweight branch
---

# Wire /agn:validate task lightweight branch

# Wire /agn:validate task lightweight branch

## Problem statement

`/agn:validate task` is a placeholder. It should be a light, main-session check that runs the task file's `## Quality gates` end-to-end, reports results, and (optionally) offers to close the task. No QA sub-agent — gates are short and the parent session has the right context.

## Scope

In scope:
- Edit `plugins/agn/skills/validate/SKILL.md`, `$0 = task` branch.
- Replace placeholder with a workflow that reads the task's `## Quality gates`, executes each, reports per-gate pass/fail, and offers `taskman.sh move <path> done` if all pass.
- Document that the QA sub-agent is intentionally not used at task level (the implementer's context is necessary for narrow gate checks).

Out of scope:
- Feature/epic/product branches (sibling task).
- QA sub-agent creation (sibling task).

## Acceptance criteria

- `$0 = task` branch has explicit workflow (no placeholder).
- Workflow runs each gate, reports pass/fail.
- On all-pass, offers (but doesn't auto-execute) the move-to-done step.
- Section explicitly notes "no QA sub-agent at this level".

## Quality gates

- Skill body parses; YAML frontmatter intact.
- Task branch has a numbered workflow, not just prose.

## Summary

### Steps completed

1. Replaced `$0 = task` placeholder in `plugins/agn/skills/validate/SKILL.md` with a 5-step workflow: Read the task → Run each gate → Report per-gate pass/fail → Offer move-to-done (do not auto-execute) → Stop.
2. Added Discipline section: no QA sub-agent at task level (gates are narrow; parent context is what matters); skill does not modify the task body; failures generate bug tasks via `/agn:define task --kind bug`, not inline fixes.
3. Preconditions handle the edge case of validating an already-`done` task (warn the user; ask if intentional).

### Changes made

Modified:
- `plugins/agn/skills/validate/SKILL.md` (`$0 = task` branch)

### Notable decisions

- **No QA sub-agent at task level — documented inline.** Future readers might assume parallel structure with feature/epic/product and try to add delegation. The Discipline section names this choice explicitly so the decision is durable.
- **Offer next step, don't auto-execute.** All-pass gates do not automatically move the task to done. The user owns lifecycle transitions. This matches the broader pattern: the validate skill reports; the user advances.
- **Failures route to bug tasks, not inline fixes.** Keeps the "no skill writes task files directly" discipline. If a gate exposes a real bug, it deserves its own task with its own scope and acceptance criteria.

### Links

- Modified file: `plugins/agn/skills/validate/SKILL.md`
- Sibling tasks: `create_qa_subagent`, `create_validate_feature_epic_product_skills`