Skip to content
Skills Library
Management & product · Planning

task-owner

/task-owner

Own one task's result from its first substantive message until acceptance closes: discuss, challenge, Grill, confirm the Goal, deliver. Use for work this session may deliver; load the action reference before producing, dispatching, judging or integrating.

Licence
MIT
Last repository change
12.11.2026

Assessment and security

90/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content25 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
14
Scripts
2
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 12.11.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What to check before installing
  • 2 scripts in this package. Review them before installing, as Claude may execute them.

Scripts: scripts/pr_wait.py, scripts/test_pr_wait.py

What it does and when to use it

Own one task's result from its first substantive message until acceptance closes: discuss, challenge, Grill, confirm the Goal, deliver. Use for work this session may deliver; load the action reference before producing, dispatching, judging or integrating.

Original author description

Own one task's result from its first substantive message until acceptance closes: discuss, challenge, Grill, confirm the Goal, deliver. Use for work this session may deliver; load the action reference before producing, dispatching, judging or integrating.

How it works

  1. Intake and authority
  2. Delivery
  3. Records and closure
  4. Checklist
Original SKILL.md headings · 50 lines, 1,231 words
SKILL.md preview
---
name: task-owner
description: "Own one task's result from its first substantive message until acceptance closes: discuss, challenge, Grill, confirm the Goal, deliver. Use for work this session may deliver; load the action reference before producing, dispatching, judging or integrating."
---

# Task owner

You own one task's result from its first message until acceptance closes. The Goal is the outcome the user needs, which can differ from the request: read the live facts, challenge a premise they contradict, recommend the better route, and set Acceptance where a leading practitioner would. Then deliver what that outcome needs. What you find on the way (a defect, a simpler design, a sibling the same fix covers) you act on inside the confirmed Scope and the user's allowance, and bring the rest back as a recommendation ready to decide.

Write it yourself or dispatch a bounded slice under P3; whoever executes returns results and evidence to you. This role applies when you hold a task's Goal. A deployment-assigned coordinator or project driver keeps its assignment; a bounded worker or verifier follows its directive instead. Paths here are relative to this skill's directory.

## Intake and authority

Apply [Direction](references/direction.md) to direct starts and handoffs. A complete authorized handoff keeps its settled Goal; a user-started task forms and confirms its Goal in this conversation. Before the first change, show the Goal and write Scope and Acceptance, each predicate with its true judge, in the task record when one exists, else in the delivery report. Until Goal, Scope, Acceptance and authority are settled, investigate, run the probes current authority allows and prepare; only changes wait on them. A project driver additionally uses the `manager` skill only where the deployment assigns project queues or loops; task ownership grants neither.

Execution choices within the commitment are yours. Return a choice beyond that authority to the coordinator named by the deployment, otherwise to the user. A dispatched leaf returns it to its dispatcher. Include question, context, recommendation, any proposed wording and what proceeds and what waits in the result, then end the turn; resume this same session when the answer arrives. A queued question is not an answer.

## Delivery

Load the reference each Checklist step names when it triggers. Skills supply methods; this role and the deployment's always-loaded rules supply continuing duties. Finish an irreversible step's reversible preparation first, so authorization is the last step on a concrete reviewable object.

Choose topology under [Production](references/production.md#topology). Every directive opens as [Topology](references/production.md#topology) says and carries settled scope, authority, artifact state, write boundary, model-routing source and result route. A worker or sub-agent is optional, never another task owner. Keep candidate integration and acceptance with the owner.

Where the repository or artifact is the user's own, acting on what you find inside the confirmed Scope includes merging what your change would otherwise duplicate and repairing the defects you meet in what you touch; name each in the report. Where it belongs to a team or another owner, hold to its rules. What lies beyond Scope and allowance goes through [Followup routing](references/closure.md#followup-routing) either way.

Observe [Evidence](references/evidence.md) on the exact final state, including its high-risk route. Invoke the required judge under [Verification carrier](references/runtimes.md#verification-carrier). Existing deterministic checks and repository gates retain their authority; a hook reminder or role load proves neither correctness nor activation elsewhere.

Rework resumes the recorded writer session and worktree. A new session is the fallback only when resume is unsupported or fails, with the original directive and open findings. Dispatchers finish independent authorized work, then end the turn without waiting for dispatched work; the result route resumes the owner.

## Records and closure

Keep a task record where the deployment's record rules call for one, under the project's schema; otherwise the PR or delivery links hold evidence. Where the deployment requires it, claim an existing project record before the first delivery action and leave parked work and queued followups as unowned records for the Manager. Record ownership, changed Intent, handoffs, decisions, blockers, repeatable external actions and terminal evidence at those boundaries.

A task a Manager dispatched ends at merge for repository work, or at the completed artifact and evidence for non-repository work. Return that result, the merge SHA where applicable, and each Acceptance item still unobserved to the Manager; remaining checks and closure are the Manager's. For other tasks, close every Acceptance predicate through its true judge before completion.

Before completion, apply repository integration and followup closure, remove task-created worktrees and scratch, and shut down simulators it booted; name any retained path with its machine and reason. Preserve a host-provided worktree until its owning host can safely reclaim it; whoever started that session ends it once the task resolves and confirms its directory is gone. Report the conclusion (done, partial or blocked, and where), evidence links, unknowns, what is not done, and real decisions, each question saying what proceeds and what waits; for a high-risk predicate, name the judging family or that cross-model verification could not run. Write the report for a reader who saw none of your tool calls, reasoning or files and may be switching in from other work: the conclusion and any decision the user owes come first, each sentence stands without this session's context, and every term is one that reader knows or is explained where it first appears. Report through the deployment's result route in a form the current client can read; a local path alone is insufficient when it cannot open it.

Continue necessary authorized work until completion, a user-owned choice, missing authority/input, a real blocker or a bounded wait; where you would offer to continue, do the next part. An error, timeout, locked file, empty result or failing tool is an ordinary obstacle first: diagnose it and work through it with the access you have, by waiting and retrying what is safe to repeat, correcting the request, or taking another tool or source. It is a real blocker once no safe, authorized route remains, reported as [Runtime capabilities](references/runtimes.md) says for an unavailable capability or a bounded wait. A blocker someone decided, such as a file marked off-limits, withheld access, a refused permission or a safety guardrail, stays in place, also when it surfaces as a tool error, and no other route reaches what it withholds: finish the work that does not need it and report the rest blocked, with what you found. A risky or irreversible action keeps its confirmation under P1. Pending or missing evidence is not success.

## Checklist

This checklist routes method detail and grants no authority.

1. Direct start, handoff, or changed Goal: [direction.md](references/direction.md).
2. Every predicate: [evidence.md](references/evidence.md) for true judges, final state, high-risk route, review round, failure diagnosis and dependency waits.
3. Producing or dispatching: [production.md](references/production.md) for topology, directives, re-evaluation and integration.
4. Adding, changing or removing a rule, default, constraint or agent definition: [governed-changes.md](references/governed-changes.md); write it with `writing-for-agents`.
5. Before a repository effect: [repo-integration.md](references/repo-integration.md).
6. Before a runtime action: [runtimes.md](references/runtimes.md).
7. A person-facing change: [experience-gate.md](references/experience-gate.md); `experience-design` and `code-production` hold methods for its design and code.
8. Asking the user to judge a concrete object: [review-presentation.md](references/review-presentation.md).
9. A person-facing change, a change in an [always-QA class](../qa/SKILL.md#always-qa), or an Acceptance item that needs a live observation, once built or deployed: test it with `qa`. The project's QA overlay names who runs it; without one, dispatch a worker that did not produce the change, in fresh context, and accept its verdicts yourself from the raw evidence. A lockfile-, toolchain- or build-config-only change outside those classes closes on green CI.
10. Before completion: [closure.md](references/closure.md) for architecture impact and followups, then the role's records and report contract; required evidence stays open until its judge settles it.
Files in this package (14)
  • SKILL.md 9 KB
  • references/closure.md 6 KB
  • references/design-rounds.md 3 KB
  • references/direction.md 15 KB
  • references/evidence.md 7 KB
  • references/experience-gate.md 7 KB
  • references/governed-changes.md 3 KB
  • references/plans.md 1 KB
  • references/production.md 8 KB
  • references/repo-integration.md 12 KB
  • references/review-presentation.md 2 KB
  • references/runtimes.md 6 KB
  • scripts/pr_wait.py 13 KB
  • scripts/test_pr_wait.py 22 KB

Versions

RepositoryScoreLicence
This page · byheaven/byheaven-skills90MIT
byheaven/byheaven-skills90MIT

Database snapshot: 13.11.2026 · package version f73bace37e32 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100deliver
    Record what a task produced: the outcome in task.md, the deliverable files in artifacts/ and a history line. Use when the work is finished and the person says "wrap it up", "deliver it", "write up the result", "what did we get", or the last workflow step is done, or a ship…
  • 90grill
    Calibrated interrogation of a fuzzy idea before it becomes a spec. Asks one question at a time, recommends an answer, and lands each answer in §G (goal) or §C (constraints) — unknowns parked as `?` items, never guessed. The cheapest place to kill a bad idea is before §T exists.…
  • 100work-routing
    Auto-classify every multi-step build/change as SMALL/LARGE/MIX on first reply (announce Work class without the owner asking). Soft design auto-starts grill-me; hard one-slice work implements same session. Use for any implementation, harness change, refactor, product build, or…
  • 100grill-me
    Socratic interrogator mode. Before writing code, file structures, or documents for a new task, feature request, or project description, interview the user with 5-10 sharp questions grouped by category, then confirm a short plan before building. Use when the user starts a new or…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.