| 1 | --- |
| 2 | name: full-send |
| 3 | description: Trigger this if the user says 'full send'. |
| 4 | --- |
| 5 | |
| 6 | # Full Send |
| 7 | |
| 8 | If the user triggered this, it's because they are not watching this thread, and |
| 9 | they want you to complete the work end-to-end autonomously. Your updates and |
| 10 | your final message go unread until long after you stop. You own the work from |
| 11 | its current state to a merged PR. |
| 12 | |
| 13 | If the user did not explicitly request "full send" in some clear way, do not |
| 14 | assume full autonomy. If they did, everything below applies. |
| 15 | |
| 16 | The user grants you permission to: |
| 17 | |
| 18 | - Merge the PR once the merge gate passes. |
| 19 | - Commit and push the task's branch under |
| 20 | [file-pr's push policy](../file-pr/SKILL.md#push-permission). |
| 21 | - Use the test methods that Proof needs, without asking first. |
| 22 | |
| 23 | ## Testing permission |
| 24 | |
| 25 | A "full send" request is the permission. If a project file or other |
| 26 | instruction says to ask before a test method, or to leave testing to the |
| 27 | user, this request replaces that rule for this task. |
| 28 | |
| 29 | That includes ask-first rules for computer use, browser automation, |
| 30 | simulators and devices, running the app or a dev server, screenshots, and |
| 31 | visual checks. Use the method that gives direct proof. List it in the |
| 32 | report. |
| 33 | |
| 34 | Rules about how to test still apply. Follow rules that name the correct |
| 35 | command, fixture, or check. This permission only removes the ask-first or |
| 36 | leave-it-to-the-user step. If a preferred method cannot prove the change, |
| 37 | use a gated method under this permission. |
| 38 | |
| 39 | It covers only the tests needed for this task. It does not add permission |
| 40 | for the steps under "Asking the user", such as production changes or data |
| 41 | deletion. It also does not bypass a harness or system approval prompt if |
| 42 | one appears. |
| 43 | |
| 44 | ## Steps |
| 45 | |
| 46 | 1. **Finish the change.** Complete the goal from the user's request with the |
| 47 | smallest change that works. Fix the case that happened, not every case |
| 48 | that could. If you are on the default branch, create a branch first. |
| 49 | 2. **Prove it.** Get proof for every change you made. See "Proof". |
| 50 | 3. **Get feedback.** Use the `sol-loop` skill to verify changes. |
| 51 | 4. **File the PR** with the `file-pr` skill. Update against the latest default |
| 52 | branch first, following its push policy for existing PRs. Add one short |
| 53 | line to the description that says how you proved the change. |
| 54 | 5. **Babysit the PR** with the `babysit-pr` skill. Fix real findings and CI |
| 55 | failures. Get proof for each fix before you push it. |
| 56 | 6. **Merge** when the merge gate passes. |
| 57 | 7. **Report** in your final message: the PR link, the merge commit, your proof, |
| 58 | and each decision you made for the user. |
| 59 | |
| 60 | Stay active from step 1 to step 7. If you end your turn while checks or bots are |
| 61 | pending, the work stops, because nobody will prompt you again. Wait with the |
| 62 | watcher from `babysit-pr`, so something wakes you when the PR changes. In T3 |
| 63 | Code that is `watch_pull_request`. If you stop before the merge, call |
| 64 | `unwatch_pull_request` first. |
| 65 | |
| 66 | ## Proof |
| 67 | |
| 68 | Proof is evidence you observed yourself in this session: command output, a |
| 69 | reproduction, a screenshot. Use the most direct proof the change allows: |
| 70 | |
| 71 | - Reproduce the original failure, apply the fix, then watch the same |
| 72 | reproduction pass. |
| 73 | - Run the checks that the repo's CI runs (typecheck, lint, tests). |
| 74 | - Add one focused test that fails without the fix, when the behavior has none |
| 75 | and the repo has a test setup. |
| 76 | - For UI and runtime behavior, run the app and look at the change. Green mocked |
| 77 | tests do not prove that the user experience works. Attach screenshots to the |
| 78 | PR when they help and you have a way to upload them. |
| 79 | |
| 80 | When you cannot find a way to prove a change, you do not have full confidence. |
| 81 | Ask the user. |
| 82 | |
| 83 | ## Merge gate |
| 84 | |
| 85 | Full confidence means all of these are true on the latest commit: |
| 86 | |
| 87 | - Your proof covers every change in the diff. |
| 88 | - Required checks are green. |
| 89 | - Review bots have finished, and each finding is fixed or dismissed with a |
| 90 | written reason. |
| 91 | - No human review still requests changes. |
| 92 | - The branch is up to date with the default branch, and GitHub reports it as |
| 93 | mergeable. |
| 94 | |
| 95 | Merge with the method the repo uses. Check the allowed methods and recent merged |
| 96 | PRs. If the repo uses a merge queue or auto-merge, stay until GitHub reports the |
| 97 | PR as merged. Merge through branch protection as configured. Never use `--admin` |
| 98 | or other bypasses. You can ignore automated reviews that request "needs human |
| 99 | review" if you have high confidence in your changes. |
| 100 | |
| 101 | ## Asking the user |
| 102 | |
| 103 | A question in a normal message goes unseen, and the thread looks done. To ask |
| 104 | the user something, use the harness's question tool so the thread shows that |
| 105 | it is waiting: `AskUserQuestion` in Claude Code, `request_user_input` in Codex, |
| 106 | or the question tool your harness has. Put all the context in the question: |
| 107 | the PR link, what blocks you, and the options with your recommendation first. |
| 108 | |
| 109 | Ask only when you cannot continue alone: |
| 110 | |
| 111 | - You cannot prove the change. |
| 112 | - The merge gate cannot pass, and you cannot fix that yourself. |
| 113 | - The next step is outside the grant above: deleting data, touching production, |
| 114 | changing things outside the repo, closing a PR, or force-pushing any other |
| 115 | branch. |
| 116 | - The goal is ambiguous in a way that changes the result. |
| 117 | - Another PR makes this one obsolete. |
| 118 | |
| 119 | Decide everything else yourself with sensible defaults, and list those decisions |
| 120 | in the report. |
| 121 | |
| 122 | If no question tool is available, or the tool returns no answer, take the safe |
| 123 | path: leave the PR open and unmerged, and start your final message with the |
| 124 | question. |