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

cc-deliver

Skill0 starsChanged 19 days ago0BSDView on GitHub

Deliver committed plan work as pull or merge requests against each repository's base branch, whether the work is one plan, one bound checkout, or a run of several plans.

  • Commits and pushes

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Several plans run together, or "deliver all changes": Ask the CLI rather than reading the dependency graph yourself:

    line 26
  2. Very specific:

    Use the exact plan_ids selected for the execution run, whether the plans are standalone or intent-linked.

    line 33
  3. Distinctive:

    Do not deliver the last wave.

    line 42
  4. Very specific:

    Read base_branch from the worktree you are delivering, or from the repository when the work is in its bound checkout.

    line 74
  5. Distinctive:

    Use ordinary Git and provider tools.

    line 82

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Deliver plan work

"Deliver the changes", "deliver all changes", and "open a PR" are requests to push committed work and open a pull or merge request for it. This skill says which branch each request opens from, and what it targets.

It carries no authorization. The request authorizes push and request creation; merging is separate and still needs a person, as do deployment and publication. Delivering does not mark a plan completed.

Which branches to open

Delivery is per repository. A plan has one worktree and one branch in each repository it names, so a plan can produce more than one request.

Work in a bound checkout. Direct execution commits to the branch that checkout is already on. Deliver that branch; there is no cc/* branch involved.

One plan. Deliver cc/<plan>/<repo> for each repository the plan names.

Several plans run together, or "deliver all changes". Ask the CLI rather than reading the dependency graph yourself:

sh <root>/.context-circuit/cli --json record order --plan p0001 --plan p0002

Use the exact plan_ids selected for the execution run, whether the plans are standalone or intent-linked. An intent-scoped run may use --intent i001 instead. An omitted filter selects all active plans and may include unrelated work.

Each plan carries deliver: the repositories whose chain ends at that plan. Open one request for every mark, from cc/<plan>/<repo>. A chain end's branch already contains the plans it was prepared from, so those plans need no request of their own — delivering them too would duplicate commits.

Do not deliver the last wave. Waves answer readiness, not chain ends: a plan nothing depends on sits in wave 1 and still ends its chain, so delivering by wave drops it silently.

Skip a branch holding no commits beyond its base. A plan can name a repository it turned out not to change, and an empty request wastes a review.

Why the marks are per repository

A predecessor that does not share a repository contributes nothing to that repository's branch, so one plan can end a chain in one repository and sit mid-chain in another:

p0003  repositories: web, sdk   depends_on: p0001   deliver: sdk
p0006  repositories: web        depends_on: p0003   deliver: web

p0003 is delivered for sdk, where nothing follows it, and not for web, where cc/p0006/web already carries it. One repository can also hold two chain ends and take two requests. Deliver exactly the marks, no more and no fewer.

Where to target

The base branch is this machine's recorded base for that repository, falling back to the repository's default branch when the binding records none:

sh <root>/.context-circuit/cli --json worktree inspect --repo api --path <worktree>
sh <root>/.context-circuit/cli --json repo inspect --id api

Read base_branch from the worktree you are delivering, or from the repository when the work is in its bound checkout. workspace.yaml records a default_branch. That is the repository's default and not what this machine delivers to; answering from it targets the wrong branch convincingly. Use base_branch, or a branch the person names.

Opening the request

Use ordinary Git and provider tools. The CLI never delivers.

git -C <worktree> push --set-upstream origin cc/p0001/api
gh pr create --base <base_branch> --head cc/p0001/api --title "..." --body "..."

Confirm the work is committed and pushed first. Uncommitted work is not delivered by a request, and a worker that reported without committing left nothing to open.

Report each request with its URL, the branch it opened from, the branch it targets, and the plans it carries. Where two chain ends share a predecessor, that predecessor's commits appear in both requests and merging either shrinks the other — say so rather than letting a reviewer meet the same commits twice.

Report drift or conflicts against the base plainly. Do not rebase, force-push, or re-verify automatically to clear them.

If the person wants one request per plan

Within a single run of plans that depend on each other and are delivered together, some teams review one request per plan, each targeting its predecessor's branch so it shows that plan's diff alone. Do that when asked, and say the requests must merge in order.

This applies only inside one such run. A plan from a separate intent, or one whose predecessor has already merged, has no predecessor branch to target: it is delivered against the base branch like any other. Most delivery is this case — one plan, finished and merged, then the next.

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.