Skip to content

gpsaggese/helpers / .claude/skills/auto_task.execute_with_stacked_prs

auto_task.execute_with_stacked_prs

Verified

Execute a sequence of tasks as one stack of branches and PRs

Goal

  • Execute one GitHub issue as a stack of sequential branches and PRs
    • Name each branch <BASE>_<ID> (ID = 1, 2, 3, ...)
    • Branch each one from the previous branch, so the PRs stack in review order
  • Create the whole stack without stopping for review
    • The user reviews, edits, and merges each PR bottom-up later
  • Follow .claude/skills/auto_task.rules.md for how tasks are queued, specified, and named

Input

  • The user passes a task in the format

When to Use This Skill

  • Use only when tasks favor stacked execution: complete specs, a real dependency chain, and a user who prefers one batch review over interruptions
  • If a spec is unclear, follow .claude/skills/auto_task.rules.md section "Ask for Clarification Before Executing an Unclear Plan". Do not build on a guess
  • Resolve all doubts with the user before you start, so the stack runs uninterrupted

Workflow

Confirm the Task List

  • Follow .claude/skills/auto_task.rules.md section "Confirm the Task List Before Executing". Each ordered task is one branch
  • Confirm task <ID+1> depends only on <ID>, not on master or earlier tasks
  • If the order or a dependency is unclear, ask before starting. A wrong dependency means rebasing the whole stack above it

Create the Issue and the First Branch (_1)

  • Follow .claude/skills/auto_task.rules.md section "One GitHub Issue Is the Unit of Work": one issue covers the whole stack

  • Create the issue and the first branch / PR <BASE>_1

    > git_create_issue_and_branch.py \
        --gh_issue_title "<TITLE>" --gh_issue_body "<Description of entire stack>" \
        --suffix 1
    
  • The description of the entire task is

  • Commit the file passed by the user (e.g., tasks.md)

  • Implement task 1, run its tests, commit, push

Stack Each Following Branch (_<ID>)

  • For ID from 2 to N, while still checked out on <BASE>_<ID-1>:
    • Derive <BASE>_<ID> from the issue, not from the current branch name. Branch it from <BASE>_<ID-1>, not from master:

      > invoke git_branch_create --issue-id <NUM> --suffix <ID> \
          --no-only-branch-from-master --no-create-pr
      
      • Do not use invoke git_branch_next_name. It appends _1, _2, ... to the current branch, so from <BASE>_<ID-1> it gives <BASE>_<ID-1>_1
      • Pass --no-only-branch-from-master. The default only_branch_from_master (True) branches from origin/master, not from the current branch
    • Implement task <ID>, run its tests, commit, and push

    • Open the PR against <BASE>_<ID-1>, not master:

      > gh pr create --base <BASE>_<ID-1> --head <BASE>_<ID> --title "<TITLE>"
      
    • Do not stop for review. Move to <BASE>_<ID+1> once <BASE>_<ID>'s tests pass

Report the Stack

  • Confirm each PR's base with gh pr view <BASE>_<ID> --json baseRefName
  • Report one ordered list to the user: branch, PR link, and base branch, for every ID

How the Stack Gets Merged

  • This is the user's process. It is documented so a later run can resume a stack
  • The user reviews, edits, and merges PRs bottom-up, one at a time
  • After <BASE>_<ID-1>'s PR merges into master:
    • Retarget <BASE>_<ID>'s PR to master, if GitHub did not do it when the merged branch was deleted: gh pr edit <BASE>_<ID> --base master

    • Sync <BASE>_<ID> with the updated master, resolve any conflicts, then commit and push:

      > invoke git_merge_master
      
      • This is a merge, not a rebase. After a squash merge, expect a conflict even on identical content: Git cannot tell the squashed commit from the originals
    • Repeat for <BASE>_<ID+1> once <BASE>_<ID>'s PR merges

  • Never force-push a branch whose PR has unresolved review comments without telling the user first

Conventions

  • Follow .claude/skills/auto_task.rules.md for queue, spec, naming, and coding and testing conventions
  • If a task needs splitting mid-run, use .claude/templates/auto_task.template.md
  • If the issue also spans a submodule (see .claude/skills/auto_task.rules.md "Multi-Repo Issues, Branches, and PRs"), only the outer repo stacks
    • The submodule gets one companion branch and PR, since invoke git_branch_create --suffix does not fan out to submodules

    • Pass --submodules with --suffix 1 on the first git_create_issue_and_branch.py call, then refresh the companion PR links after the stack is built:

      > git_create_issue_and_branch.py --gh_issue_id <NUM> --submodules \
          --update_pr_links
      

Constraints

  • Follow .claude/skills/auto_task.rules.md sections "The User Owns Commit and Merge Decisions" and "One GitHub Issue Is the Unit of Work"
  • Do not start the next task before the current task's tests pass. An untested task compounds into everything stacked on it
  • Do not squash or reorder commits across tasks unless asked
  • Keep one task per branch and one PR per task. Do not fold queued tasks together
  • One GH issue covers the whole stack. Do not open an issue per branch

Verification

  • Every branch is named <BASE>_<ID>, with ID rising by 1 from <BASE>_1
  • Each branch's merge base is the previous branch, and <BASE>_1's is master
  • Each PR's base is its parent branch, and <BASE>_1's PR base is master
  • Every branch and PR sits under the one GH issue created with <BASE>_1
  • Each task's tests passed before its branch was pushed
  • The final report lists every branch and PR in dependency order, with links
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 192kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 192kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 192kUpdated Claude CodeCodexVerifiedInstall