agentleFS
kaotyprkaotypr/context-circuit/.agents/skills/cc-direct/SKILL.md

cc-direct

Skill0 starsChanged 17 days ago0BSDView on GitHub

Make a change directly in a bound checkout when a person chooses to bypass intent and planning — what their request authorizes, where the work happens, when to stop and offer the intent path, and how the knowledge it changed is reconciled.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 4 are distinctive, and 1 is 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.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Never checkout, reset, stash, or force anything to make the change fit.

    line 31
  2. Distinctive:

    Leave the change uncommitted unless the person authorizes a commit.

    line 35
  3. Distinctive:

    Run the checks the change deserves, and report what was actually run and what it actually returned.

    line 39
  4. Distinctive:

    Leave what is already correct in place, and say plainly what is done and what is not.

    line 54
  5. Very specific:

    Use .agents/skills/cc-knowledge/SKILL.md for the writing itself.

    line 73

≈ “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

Change it directly

A person has chosen to bypass the intent and plan path for this change. Work in the bound checkout, on the branch it is already on, and keep the change to what they asked for.

What the request authorizes

The request is the approval. The intent gate exists because an agent's reading of a request can drift from what the person meant, so the outcome is written down and approved before code is investigated. Here the outcome is what they said: no reading was derived, so there is nothing derived to approve. That is the whole of the bypass, and it holds only because a person asked for it.

It authorizes changing what the request describes and nothing beyond it. No record is written, no intent or plan ID is allocated, and nothing is reserved or archived. Every other gate stands exactly as it does elsewhere: committing in a bound checkout, pushing, opening a pull request, merging, deploying, publishing, and deleting each still need explicit authorization.

Where the work happens

The bound checkout this machine records, on its current branch. No worktree is prepared and no branch is created, because isolation is what the person declined.

  • Never checkout, reset, stash, or force anything to make the change fit. Work with the tree as it is.
  • Preserve everything already there — staged, dirty, untracked, and ignored. Uncommitted work belonging to someone else is not yours to tidy.
  • Leave the change uncommitted unless the person authorizes a commit. The diff in front of them is what a plan record and a branch would otherwise have been.
  • The repository's own AGENTS.md or CLAUDE.md applies here exactly as it does in a prepared worktree; follow it where it is more specific.
  • Run the checks the change deserves, and report what was actually run and what it actually returned.

When to stop and offer the intent path

The lane fits a change whose outcome is already settled. Stop as soon as it is not, say what was found, and offer to write an intent instead:

  • the change reaches a second repository, or has to be coordinated with work already planned;
  • settling it needs a decision the person has not made;
  • the fix reveals the real problem is somewhere else;
  • the reading needed to make the change safely is large enough to have been a plan.

Leave what is already correct in place, and say plainly what is done and what is not. A change that grew past this lane and was carried through anyway is an unrecorded feature nobody approved.

Reconcile the knowledge it changed

A completion names this work for a plan. A change made without a plan has no record to name it, so ask the repository:

sh <root>/.context-circuit/cli context find --repo REPO

It returns the catalog entries claiming that repository and names the judging as reconcile_required. Judge each against what the change actually altered: edit the note and its catalog entry together and move its reviewed date, or say in your report that it changed nothing. Most direct changes change nothing durable, and saying so is the normal outcome rather than a skipped step.

Use .agents/skills/cc-knowledge/SKILL.md for the writing itself. This is the one trace a direct change leaves besides the diff, which is why it is not optional.

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.