# Task

> Stateless end-to-end build loop: grill, plan, implement, gather acceptance evidence, and review one verifiable outcome using in-chat execution context. Use when the user wants a feature or outcome built end to end. Nested under /just-do-it when that parent owns shipping.

- Skill: `beskars-code/task` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add beskars-code/task`
- Raw SKILL.md: https://api.skillmd.com/api/skills/beskars-code/task/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: beskars-code (https://skillmd.com/u/beskars-code)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/beskars-code/task

---


# Task

**Must read:** [../pack-shared/standards.md](../pack-shared/standards.md) — Read `/taste` and `/architecture` doctrines this turn before grilling, planning, or implementing. Do not skip.

**Execution context:** [../pack-shared/execution-context.md](../pack-shared/execution-context.md) · **Read:** [doctrine.md](doctrine.md) · [reference.md](reference.md) · **Ask style:** [../pack-shared/asking.md](../pack-shared/asking.md) · **PR ship:** [../pack-shared/pr-ship.md](../pack-shared/pr-ship.md) when this chat opens a PR

Orchestrator only. Dispatch Task workers per
[../pack-shared/subagents.md](../pack-shared/subagents.md) and review their
Completions; do not solo non-trivial labor or grep the tree. Always load
`/taste` and `/architecture` during grill and before every implement wave.
Load `/design` before every user-facing implement wave.

`/task` follows the stateless
[execution-context contract](../pack-shared/execution-context.md): plans are
inline unless the user explicitly requests a saved artifact and approves its
destination.

## Process

1. Establish or refresh the in-chat execution context.
   If a parent already supplied ticket, lane, Done when, non-goals, Active
   Rules, fixed point, and slice bounds, accept that brief. Do not re-derive
   ticket or branch ownership the parent holds.
2. Run the [lifecycle](reference.md#lifecycle): grill (unless skip-grill
   applies) → plan → implement → acceptance evidence → `/code-review` → Fix
   mode as needed. Pick Task specialists per
   [subagents.md](../pack-shared/subagents.md): the catalog, not a fixed
   spawn order. Main does not grep or write tests. Always
   load `/taste` and `/architecture` during grill and before every implement
   wave. Load `/design` before every user-facing implement wave.
3. Announce completion.

### If a parent already owns the ticket, branch, and PR

Do not ask ship Questions. Do not commit, push, or open a PR from this skill.
Return a completion summary plus evidence envelope to the parent.

### If this chat owns shipping

Offer ship Questions only after all gates pass (see
[reference.md](reference.md#ship-questions)). Do not commit or open a PR
unless the user answers yes. If they ask to open a PR, follow
[../pack-shared/pr-ship.md](../pack-shared/pr-ship.md). Do not invent a parent.

Recovery, progress, lookup, and safety rules live in the doctrine and reference.

## Anti-patterns

- Asking ship Questions when a parent owns shipping
- Pushing, committing for ship, or opening a PR when nested

