Claude Skill

milestone-sweep

Runs verify, triage, next-milestone, fix and release-gate across every philoserf repo, one background agent per repo in its own worktree, and stops before shipping. Use when asked to sweep, groom or get the philoserf repos release-ready. Holds every decision for one batch at the

LLM Mart · 0 points · 2 views 0 listing impressions 0 install-command copies
Download philoserf-claude-code-config-skills_milestone-sweep-b075299.zip · 2 KB

Install

skills CLI npx skills add https://github.com/philoserf/claude-code-config/tree/main/skills/milestone-sweep
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install philoserf-claude-code-config@llmmart
Git git clone https://github.com/philoserf/claude-code-config.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole philoserf/claude-code-config collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Milestone sweep

One agent per repo under ~/source/philoserf/, each working its repo end to end in a linked worktree: verify, triage, pick the next milestone, ship it with issue-pr, gate it with release-gate, stop before tagging. You orchestrate and do not touch the repos yourself while their agents run.

Run it on almanac, from a session started in ~/source/philoserf. almanac is the always-on host. Twenty agents report back over an hour or more, and nothing here needs sextant. Starting anywhere else, such as ~/notes, roots the language server outside the repos. It then sends a stream of false "module not in workspace" and undefined: errors about files that compile.

1. Sync, then list the repos

cd ~/source/philoserf && task sync

Stop if sync reports a repo as diverged or errored. Don't sweep a clone that doesn't match its remote.

The repos are the child directories with a .git. With arguments, sweep only those. Leave out:

  • obsidian-vault-changelog, excluded from all workspace work (CLAUDE.md)
  • any repo whose gh repo view fails, because there is nothing to triage or open PRs against

Note any repo with issues disabled (gh repo view --json hasIssuesEnabled). Its agent runs only step 5, the gate, and lands any fix as a PR with no issue.

2. Launch

One Agent call per repo, all in one message, with subagent_type: "repo-sweeper"; agents run in the background on their own. The agent definition (~/.claude/agents/repo-sweeper.md) carries the whole brief. The prompt only names REPO and SCRATCH, where SCRATCH is your scratchpad directory, plus anything repo-specific you know: that it is one of the unversioned sites, that issues are disabled, or that a milestone already exists.

Tell the user how many agents went out, which repos you skipped, and why.

3. While agents report

Relay each report as it lands, briefly. Assume the user is not watching. A relay is a heads-up, not a question that blocks anything. Nothing waits on an answer: agents keep working, and every item the user hasn't answered carries into step 4. Do not act on a report's requests yourself:

  • An agent blocked by a permission prompt reports the command it needed. That command goes to the user. You don't run it.
  • A pending decision is not answered until the user answers it. If they answer mid-run, act on it. If not, it waits for the batch.

Do act on a fleet-wide hazard. If one agent reports a mistake the others are about to repeat, send the running agents a warning with SendMessage.

4. Decisions, in one batch

Once every agent has reported, collect every decision they held back and every blocked command, leaving out what the user already answered mid-run. Present them together, with AskUserQuestion where the choices are discrete. If the user is still away, the final report lists them as open, and the run is complete without them. Typical ones:

  • a PR held because its merge would deploy
  • a kind:decision issue
  • whether a repo is versioned at all

Act on the answers.

5. Report

Per repo:

  • the milestone and its proposed version
  • what was fixed, what is blocked, and what needs a decision
  • what was deferred, and why
  • the gate result

Then list the toolchain migrations that apply across all repos.

Gate results: exit 3 with 0 FAIL and 0 WARN means nothing is ready to release, and counts as clean. Report it apart from a true exit 0.

Then offer three follow-ups:

  • /release-sweep for the repos the gate left ready to release.
  • NEXT.md rows. A row is an action, never a status. Follow ~/source/philoserf/CLAUDE.md. The file is tracked in the dotfiles repo; commit it with dot only when asked.
  • A co-log line in today's daily note.

Do not use when

  • The repos are already gated and ready — use /release-sweep, which ships them. This skill stops before tagging.
  • One issue needs fixing — use issue-pr.
  • One repo's issues need sorting, not the fleet's — use issue-triage there.
Files (claude-code-config)
  • SKILL.md 4.3 KB
    ---
    argument-hint: "[repo...]"
    description: Runs verify, triage, next-milestone, fix and release-gate across every philoserf repo, one background agent per repo in its own worktree, and stops before shipping. Use when asked to sweep, groom or get the philoserf repos release-ready. Holds every decision for one batch at the end.
    allowed-tools:
      - Agent
      - Bash
      - Read
      - Edit
      - AskUserQuestion
    ---
    
    # Milestone sweep
    
    One agent per repo under `~/source/philoserf/`, each working its repo end to end in a linked
    worktree: verify, triage, pick the next milestone, ship it with `issue-pr`, gate it with
    `release-gate`, stop before tagging. You orchestrate and do not touch the repos yourself
    while their agents run.
    
    **Run it on `almanac`, from a session started in `~/source/philoserf`.** `almanac` is the
    always-on host. Twenty agents report back over an hour or more, and nothing here needs
    `sextant`. Starting anywhere else, such as `~/notes`, roots the language server outside the
    repos. It then sends a stream of false "module not in workspace" and `undefined:` errors
    about files that compile.
    
    ## 1. Sync, then list the repos
    
    ```bash
    cd ~/source/philoserf && task sync
    ```
    
    Stop if `sync` reports a repo as diverged or errored. Don't sweep a clone that doesn't match
    its remote.
    
    The repos are the child directories with a `.git`. With arguments, sweep only those.
    Leave out:
    
    - `obsidian-vault-changelog`, excluded from all workspace work (`CLAUDE.md`)
    - any repo whose `gh repo view` fails, because there is nothing to triage or open PRs against
    
    **Note any repo with issues disabled** (`gh repo view --json hasIssuesEnabled`). Its agent
    runs only step 5, the gate, and lands any fix as a PR with no issue.
    
    ## 2. Launch
    
    One `Agent` call per repo, all in one message, with `subagent_type: "repo-sweeper"`; agents
    run in the background on their own. The agent definition (`~/.claude/agents/repo-sweeper.md`) carries
    the whole brief. The prompt only names `REPO` and `SCRATCH`, where `SCRATCH` is your
    scratchpad directory, plus anything repo-specific you know: that it is one of the
    unversioned sites, that issues are disabled, or that a milestone already exists.
    
    Tell the user how many agents went out, which repos you skipped, and why.
    
    ## 3. While agents report
    
    Relay each report as it lands, briefly. **Assume the user is not watching.** A relay is a
    heads-up, not a question that blocks anything. Nothing waits on an answer: agents keep
    working, and every item the user hasn't answered carries into step 4. Do not act on a
    report's requests yourself:
    
    - An agent blocked by a permission prompt reports the command it needed. That command goes to
      the user. You don't run it.
    - A pending decision is not answered until the user answers it. If they answer mid-run, act
      on it. If not, it waits for the batch.
    
    **Do act on a fleet-wide hazard.** If one agent reports a mistake the others are about to
    repeat, send the running agents a warning with `SendMessage`.
    
    ## 4. Decisions, in one batch
    
    Once every agent has reported, collect every decision they held back and every blocked
    command, leaving out what the user already answered mid-run. Present them together, with
    `AskUserQuestion` where the choices are discrete. If the user is still away, the final
    report lists them as open, and the run is complete without them. Typical ones:
    
    - a PR held because its merge would deploy
    - a `kind:decision` issue
    - whether a repo is versioned at all
    
    Act on the answers.
    
    ## 5. Report
    
    Per repo:
    
    - the milestone and its proposed version
    - what was fixed, what is blocked, and what needs a decision
    - what was deferred, and why
    - the gate result
    
    Then list the toolchain migrations that apply across all repos.
    
    **Gate results:** exit 3 with 0 FAIL and 0 WARN means nothing is ready to release, and counts
    as clean. Report it apart from a true exit 0.
    
    Then offer three follow-ups:
    
    - **`/release-sweep`** for the repos the gate left ready to release.
    - **`NEXT.md` rows.** A row is an action, never a status. Follow `~/source/philoserf/CLAUDE.md`.
      The file is tracked in the dotfiles repo; commit it with `dot` only when asked.
    - **A co-log line in today's daily note.**
    
    ## Do not use when
    
    - The repos are already gated and ready — use `/release-sweep`, which ships them. This skill
      stops before tagging.
    - One issue needs fixing — use `issue-pr`.
    - One repo's issues need sorting, not the fleet's — use `issue-triage` there.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.