| name | source-command-autopilot |
| description | Intent-to-outcome front door for raw context, route choice, safe workspace-local execution, proof, and risk-gated judgment |
source-command-autopilot
Use this skill when the user asks to run /autopilot, drops raw thoughts, wants
the system to choose the best workflow/skill stack, wants fewer manual passes,
needs route traceability, or wants the Antigravity harness to execute safe local
work and report proof.
Autopilot is now safe execute-by-default for Codex Antigravity workspace-local
work. It must emit Intent Lock, Autopilot Trace, Execution Decision, Chosen Path,
Capability Graph, Execution Plan, Run Prompt, Run Receipt, Friction Ledger, and
Approval/Risk Gate. The Execution Decision is one of Running now, Needs judgment, Blocked by risk, or Plan only.
Pause only for real risk or judgment gates: external/public action, publishing,
outreach, paid or budget-sensitive tools, destructive edits, global ~/.codex
changes, /Users/farricecain/Google Antigravity edits, connector writes, real
Codex subagents without explicit delegation, low-clarity execution, or explicit
--plan/--menu mode.
When blocked, include a copy-paste Run Prompt. Plugin packaging requests must go
through /plugin-readiness-audit and the helper/script -> workflow -> skill
wrapper -> plugin candidate -> plugin ladder before a plugin is recommended.
Operator Core Alignment
This project wrapper follows .agent/workflows/autopilot.md as the canonical behavior source.
It must preserve 3 Next Prompts, Operator Lesson,
Next-time prompt, Subagent worth it?, Reuse hook, and the rule that real
Codex subagents require explicit authorization.
Command Template
Read and execute the workflow at .agent/workflows/autopilot.md - Intent-to-outcome front door for route choice, safe execution, proof, and risk-gated judgment.
