autonomy-grantlisted

# /autonomy-grant ## Host dispatch contract - Codex: `$skill`, Codex collaboration roles, and GPT-5.6 tiers. - Claude: `/skill`, Agent/Skill tools, and Claude aliases. - A bare `/skill` in this shared source denotes the Claude form; Codex dispatches the same named skill as `$skill`. The operator approves the run's gate list ONCE, up front. The loop then checks the ledger mechanically instead of stopping to ask. A gate NOT in the ledger still stops — that is the safety floor, not a bug. Ported in spirit from the fable-agent-orchestration `autonomous-finish-loop` skill (Apache-2.0): proceed on approved work; hold only at genuinely unapproved external gates. ## When to run - At plan approval, before an `--autonomous` `/feature-implement` run. - Before any overnight/operator-absent run that will push, deploy, or merge. - Morning after: to approve `pending` actions a run stopped on, then resume. ## Procedure 1. **Enumerate the gates.** Walk the plan/tasks and list EVERY operator-gated action the run will perform. Typed form only — `type:target` — because free prose never matches at run time: | Type | Example | |---|---| | `push` | `push:origin/main` | | `merge` | `merge:pr` | | `deploy` | `deploy:vercel-web` | | `flip` | `flip:HOLALUMINA_301_TO_GETGLANCE` | | `restart` | `restart:tenant-gateway` | | `secret-use` | `secret-use:RAILWAY_API_TOKEN` | | `wall-reset` | `wall-reset:04-b3-specs` (target = phase-slug, `basename` of the phase dir; o