agentleFS
nshopiknshopik/dream-team/skills/assemble-meta/SKILL.md

assemble-meta

Skill2 starsChanged 19 days agoMITView on GitHub

Takes a tracker Meta issue to one MR/PR: triages the Meta issue and every child before any code, asking all open questions in one batch, builds each child up to the gate on one integration branch through the assemble-meta-loop workflow, reviews and fixes the whole branch once, and opens one MR/PR with one commit per change. Use when the user runs /dream-team:assemble-meta or asks to build, work or ship a Meta issue's children together, on one branch or as one MR/PR — e.g. "assemble-meta #40", "do Meta #40 in one MR". Not for a single issue (run build-issue).

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 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.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    No child left → report it and stop.

    line 31
  2. Distinctive:

    Run build-issue step 3 for the Meta issue: its worktree branch is the integration branch.

    line 44
  3. Distinctive:

    Run build-issue steps 6 and 7.

    line 89

≈ “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

assemble-meta

One Meta issue per invocation. Read <base directory>/../build-issue/SKILL.md first: each "build-issue step N" below runs as written there, with the Meta issue's number as <n>.

You are the orchestrator, as in build-issue: you do not write the change. Your only edits are build-issue's step-5 edits and a step-3 decision record with the files repo-conventions.md adds with it. Lab work is yours, per build-issue.

1. Read the Meta issue and its children

  • Fetch the Meta issue with build-issue step 1's fetch command; skip its Meta child pick.
  • Take every unticked link to an open child issue in its checklist; skip plain-text lines.
  • Fetch each child the same way, and note its Blocked by issues.
  • A child blocked by an open issue that is not a child → leave it and its dependents out, and report them.
  • A child whose title starts Research: → leave it and its dependents out, and report them.
  • No child left → report it and stop.

2. Triage everything before code

  • Triage the Meta issue and every child per build-issue step 2, before step 3.
  • Collect every open question first, then ask them in one batch: one AskUserQuestion call, or several in the same turn past its 4-question limit.
  • Questions whose options are not enumerable → list them all in one report and stop.
  • Handle each answer per build-issue step 2, on the issue it settles.
  • Skip build-issue step 2a: every child runs the workflow.

3. Branch

  • Run build-issue step 3 for the Meta issue: its worktree branch is the integration branch.
  • Its recorded HEAD is the workflow's base.

4. Run the workflow

Invoke the Workflow tool with name: "dream-team:assemble-meta-loop" and args as a real JSON object:

{
  "meta": { "issue": "40", "title": "<Meta title>", "body": "<Meta description, verbatim>" },
  "children": [
    { "issue": "41", "title": "<title>", "body": "<description, verbatim>",
      "notes": "<answers for it>", "blockedBy": [] },
    { "issue": "42", "title": "<title>", "body": "<description, verbatim>", "blockedBy": ["41"] }
  ],
  "notes": "<what applies to every child>",
  "branch": "issue-40-slug",
  "base": "<sha from step 3>",
  "gateCommands": ["cargo test --workspace"],
  "externalReview": "<command, or leave out>"
}
  • children: every child step 1 kept; blockedBy lists only other children's numbers.
  • A child's notes: its tracker decisions and the step-2 answers that settle it. Optional.
  • notes, gateCommands and externalReview: per build-issue step 4.
  • Then report per build-issue step 4.

5. Judge the result

  • children — each finished child's gated run: assumptions, redEvidence, degraded, untrackedPaths.
  • failed — each child whose run did not pass the gate, with its stage and reason; its commits and untracked files are reset off the branch.
  • skipped — children not run because a failed child blocks them.
  • review — the one build-issue-loop result for the whole branch, from Simplify on, with the Meta issue as its issue.
  • review is null → no child passed the gate, or stage: 'reset' left the branch at the wrong head or with files left over: hand the branch back.

Otherwise run build-issue step 5 on review, its lab work once on the branch head.

6. Open the MR/PR

  • Run build-issue steps 6 and 7; the fold leaves one commit per change.
  • Link Closes #<child> for each finished child, in place of build-issue's Closes #<n>.
  • Link Closes #<meta> only when failed and skipped are empty and step 1 left no child out.

Report back

Per build-issue's Report back, plus these lines, which count as named there:

  • Each finished child's assumptions that build-issue's Report back would show, under **Assumptions:**, prefixed with #<child>.
  • review's questions → build-issue's **Questions:** line.
  • Each failed child → one **Failed:** bullet: the child, its stage and reason.
  • Each skipped or step-1 left-out child → one **Skipped:** bullet: the child and its open blocker or Research: title.

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.