auto

Plugin auto in shawnroos/auto. Commands read from the repo's manifest — 1 month ago.

Skills (10)

• auto-author-goal

  Turn a plan into a model-judgeable goal doc the user binds with `/goal <doc-path.md>`. Use when the user says "generate a goal for this plan", "make a goal doc", "write a /goal for <plan>", "write a goal doc for this run", or wants auto to phrase an exit predicate from a plan. This skill AUTHORS the

• auto-author-workflow

  Author a new auto workflow (formerly auto-author-recipe) — the named, file-backed declaration of a loop topology — from a plain-language description, or reverse-derive one from a completed run. Use when the user says "save this as a workflow", "I want an auto workflow that does X", "make a workflow

• auto-backend

  Author an auto backend (formerly auto-adapter) — the thin shim that maps one concrete workflow (native Claude, Compound Engineering, slate-devs, plain /plan, etc.) onto the engine's six fixed operations. Use when adding a new backend to auto, when implementing the native or CE backend (U6b), or when

• auto-design

  Design a loop from the current session — turn intent into a sharp goal plus typed, checkable verification, then compile it to artifacts auto runs unchanged. Use when the user says "design a loop", "structure this run", "help me shape a goal and verification for this", "what should the done-condition

• auto-driver

  Orient before /auto starts. Loads the hypothesis JSON from lib/auto-detect.sh, surfaces one action line, dispatches when ambiguity is null, or asks one blocking question when it isn't. No verdict-tree enumeration, no workflow-picker prose. Hands off to lib/auto.sh (single plan) or lib/auto-spawn.py

• auto-launch

  The loop-design launch agent for interactive `/auto`. Assesses session + repo, recommends a fitting built-in shape (a1/a2/a4/w) or composes a custom loop, proposes typed verification gates, and drives the skip / confirm / two-step chooser before dispatch. Use when the interactive launch path needs a

• auto-preset

  Run one named preset one-shot against a target — the Phase-1 headline of addressable step presets. Use when the operator wants to fire a single tuned step (a tuned review, a scoped build) standalone: "run <preset> against <target>", "one-shot the tuned-review on this diff", "fire scoped-build on <ta

• auto-translate

  Translate a designed loop (from auto-design) or a workflow (from auto-author-workflow) into an execution tree — ordered parallel waves sized from the workflow's depends_on DAG and the active fan-out cap, fan-out do_step children nested under their producer parent, and a substrate routing decision (n

• auto-watch

  Render the live auto agent tree — driver → work step → do_step fan-out agent — with each node's age against its stall threshold and its attempt count, then overlay live-agent status from the harness. Use when the driver or operator wants to SEE what a run has dispatched, what is nested under a fan-o

• auto

  Drive an auto run: chain the plan-loop → handoff → work-loop using the self-pacing pulse (lib/pulse.py), the agent-managed dispatcher (lib/dispatcher.py), and a deliberate-stop /goal binding. Use when invoked via /auto, when continuing after a handoff, or when resuming a run. This skill IS the drivi

Slash commands (5)

• /auto:auto-pulse
• /auto:auto-resume
• /auto:auto-status
• /auto:auto-tick
• /auto:auto