Full Send
A Skill workflow for AI coding agents that, after explicit user authorization, lets the agent complete a task autonomously up to merging the pull request, including a merge gate and proof-verification mechanisms.
Timeline (last 90 days)
t3.gg places three Skills (Soul Loop, Full Send, Full Stop) in a repository called Fli, managed in folders by type, to keep authentication, MCP configuration and skills synchronized across multiple machines.
The Full Send skill is triggered only when the user explicitly requests it, indicating that the user is not watching the thread and wants the Agent to complete the work autonomously from start to finish.
Full Send's merge gate requires: proof of coverage for every change, all required checks green, all review-bot issues fixed or dismissed, no human reviewer requesting changes, the branch synchronized with the default branch, and GitHub reporting the PR as mergeable.
Full Send's proof mechanism requires the model to verify the change itself, for example by reproducing the original failure, applying the fix, and re-running the reproduction to confirm it passes; running the repository's CI checks; writing targeted tests for fixes that lack tests; or running the app to verify UI and runtime behavior.
Full Send's question gate requires the model, when the user is not watching the thread, to use the Harness's question tool (Ask User Question in Claude Code, Request User Input in Codex) rather than asking in an ordinary message.
t3.gg used Full Send 32 times within a limited time window, of which 27 were merged autonomously by the Agent without him looking again.
The Full Stop skill is almost identical to Full Send, but requires the Agent to stay responsible until the PR is ready for the user to merge and never to merge on its own.
All Facts (7)
Full Send's proof mechanism requires the model to verify the change itself, for example by reproducing the original failure, applying the fix, and re-running the reproduction to confirm it passes; running the repository's CI checks; writing targeted tests for fixes that lack tests; or running the app to verify UI and runtime behavior.
50%UnverifiedThe Full Send skill is triggered only when the user explicitly requests it, indicating that the user is not watching the thread and wants the Agent to complete the work autonomously from start to finish.
50%UnverifiedFull Send's merge gate requires: proof of coverage for every change, all required checks green, all review-bot issues fixed or dismissed, no human reviewer requesting changes, the branch synchronized with the default branch, and GitHub reporting the PR as mergeable.
50%Unverifiedt3.gg places three Skills (Soul Loop, Full Send, Full Stop) in a repository called Fli, managed in folders by type, to keep authentication, MCP configuration and skills synchronized across multiple machines.
50%UnverifiedFull Send's question gate requires the model, when the user is not watching the thread, to use the Harness's question tool (Ask User Question in Claude Code, Request User Input in Codex) rather than asking in an ordinary message.
50%UnverifiedThe Full Stop skill is almost identical to Full Send, but requires the Agent to stay responsible until the PR is ready for the user to merge and never to merge on its own.
50%Unverifiedt3.gg used Full Send 32 times within a limited time window, of which 27 were merged autonomously by the Agent without him looking again.
50%