Skip to content
Filip Stack
Documentation: full-send

full-send

Finishes a change or PR through review and CI, then merges or returns a verdict. Needs sol-loop and babysit-pr. Adapted from tstack.

Install
On its own
Runs
Manual · /full-send
Source
SKILL.md
Adapted from
full-send by Theo Browne
npx skills add filipgutica/filip-stack -g -a claude-code codex -s full-send -y

What it does

One Sol reviewer checks the code and PR description through sol-loop. The agent then opens or updates the PR, handles CI and bot findings, and makes routine decisions while you are away. For a full council, use merge-council; full send can also finish its open items.

When to use it

Use full send for a local change or existing PR you want finished, including merging when its mode and checks permit. To keep the merge decision, use full-stop.

Install full-send, sol-loop, and babysit-pr together: replace -s full-send in the command above with -s full-send sol-loop babysit-pr. It stops if either dependency is missing.

Invoke it explicitly with /full-send in Claude or $full-send in Codex. Include a goal or PR link, for example /full-send fix the flaky login test.

Calling it authorizes commits and pushes to the PR's branch in fix mode, --force-with-lease after a rebase of that branch, and merging in merge mode.

How it works

The agent selects two modes at the start and checks them again before finishing:

Mode Applies when Result
Fix You authored the PR, or it does not exist yet, and you can push to its head repository Commit and push fixes on its branch
Review only Fix mode does not apply Report findings without pushing
Merge Fix mode applies, you can push to the base repository, and the review decision permits it Merge after the full gate passes
Verdict Merge mode does not apply Return PASS or FAIL without merging

Proof must come from evidence the agent observes in the session: a failing then passing reproduction, repository CI checks, or the running app for UI and runtime behavior. Mocked tests alone do not prove the user experience.

Routine fixes stay in the loop. Public-contract, migration, data, security, and dependency changes need your decision. So does removing a test without proof that a named remaining test covers it. The agent also asks when proof is unavailable, the goal is materially ambiguous, or the next action exceeds the grant, such as deleting data or touching production. It uses the host's question tool; without one or an answer, it leaves the PR open and unmerged.

The final gate requires proof for the whole diff, Sol approval, resolved merge-council items, green required checks, completed bot reviews, no human changes request, and an up-to-date, mergeable branch. It never bypasses these checks with --admin or another override.

SKILL.md defines the grant, mode selection, proof, and complete merge gate.

What you get

  • Merged: <merge commit>, PASS: ready for human review, or FAIL: <main reason>.
  • The PR link, proof, Sol rounds, fixed and declined findings with reasons, and how each earlier merge-council item closed.
  • Decisions made on your behalf and anything you still need to do.
  • full-stop

    Finishes a change or PR through review and CI and leaves merging to you. Needs full-send, sol-loop, and babysit-pr. Adapted from tstack.

    Standalone
  • sol-loop

    Checks a change through GPT Sol review rounds and resolves findings within scope. Adapted from tstack.

    Standalone
  • babysit-pr

    Monitors a PR through CI and automated review, fixing verified problems within its scope. Adapted from tstack.

    Standalone
  • merge-council

    Runs a council on a PR, fixes agreed findings, and returns a verdict tied to the reviewed commits. Needs council.

    Standalone