agentleFS
thunderockthunderock/thunderkit/skills/tk-quick/SKILL.md

tk-quick

Skill2 starsChanged 29 days agoMITView on GitHub

Use when a task is small but not trivial: run it on one model (the planner class from .thunderkit/config.json, or model=<key> from models.json) with no plan document or parallel lanes, atomic commits and tests, and exactly one reviewer from a different model family before each commit.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    Default author: the classes.planner model from .thunderkit/config.json.

    line 20
  2. Distinctive:

    When the user must choose, use the host's structured choice tool (options as buttons, the recommended one first, plus free text).

    line 27
  3. Very specific:

    Use it when the task needs a handful of files and one owner but is more than a trivial edit.

    line 32
  4. Very specific:

    If the task needs parallel lanes, a design decision or a plan review, route it to tk-plan through tk-router.

    line 33
  5. Distinctive:

    Run the relevant tests before each commit.

    line 54
  6. Distinctive:

    Commit only after the review returns.

    line 56
  7. Distinctive:

    Only that separate controller has the scoped checkpoint exception.

    line 63

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

tk-quick: a small task on one model, with one review

tk-quick sits between tk-fast and the full lifecycle. It runs a small task on one chosen model, with atomic commits and tests, and gets exactly one review from a model of a different family before each commit. There is no plan document and there are no parallel lanes.

Model choice

  • Default author: the classes.planner model from .thunderkit/config.json.
  • Override: model=<key>, where <key> must be a key in references/models.json. An unknown key is rejected; never guess a nearby model.
  • Reviewer: one model from classes.reviewers whose family in references/models.json differs from the author's. If no configured reviewer has a different family, stop as blocked; never review with the same family.

When the user must choose, use the host's structured choice tool (options as buttons, the recommended one first, plus free text); fall back to a numbered list only when the host has none.

Scope

Use it when the task needs a handful of files and one owner but is more than a trivial edit. If the task needs parallel lanes, a design decision or a plan review, route it to tk-plan through tk-router.

Delegation

tk-quick delegates nothing (thunderkit-delegates: none). GSD quick mode requires a GSD project (.planning/ROADMAP.md), so it is not a target. After validating the configuration the resolver returns owned:

python3 scripts/tk-resolve.py --skill tk-quick --operation quick --config .thunderkit/config.json --json

That result proves only that the route is owned; it is not evidence the task ran.

Fallback

The owned procedure is the skill:

  1. Confirm the author model (default or model=) and one different-family reviewer.
  2. Make the change on the author model; keep each commit to one logical step.
  3. Run the relevant tests before each commit.
  4. Give the reviewer the diff and the test output; address findings or record why not.
  5. Commit only after the review returns. The edit stage permits no push, PR, tag or publish.

Without a valid configuration, stop as blocked and ask for one; never pick a model silently.

Save current context and the change's test/review artifacts at preservation boundaries, then have the approved checkpoint controller run checkpoint_cli.py capture under the shared preservation rule. Only that separate controller has the scoped checkpoint exception; native workers remain delivery-disabled. A snapshot does not replace the different-family review before each commit.

Asking the user

When this skill needs a decision from the user, ask through the host's structured choice tool as described in references/asking.md: one decision per question, two to four options with the recommended one first, free text always accepted. Use the numbered-list fallback only when the host has no such tool; in a non-interactive run record unknown and stop at the gate.

Output contract

task: <one sentence>
author: <model key> (<family>)
reviewer: <model key> (<family>)
commits: <sha subject>...
tests: <command> -> <exit code>
review: approved | changes addressed | blocked: <reason>

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.