Autonomy layerv0.1.0

watch-and-act

Bridges a run to external state with bounded, cache-aware polling and explicit transition handling.

WORKFLOW / 05/watch-and-act <signal to watch — e.g. 'CI on branch X' or 'deploy status'> [until <condition/deadline>]
  1. 01

    Confirm the signal is external, unresolved, and not harness-tracked.

  2. 02

    Set an absolute deadline, poll cap, and cache-aware cadence.

  3. 03

    Persist the bound and count before the first scheduled wakeup.

  4. 04

    Read the real state, classify it, and act on the observed transition.

  5. 05

    Stop on target, terminal failure, or timeout and report the final state.

operating envelope

Use it deliberately.

The shortest useful definition of where this skill helps—and where it should stay out of the way.

Use when

  • CI, a deployment, queue, endpoint, or external approval changes outside the harness.
  • The signal is unresolved and worth checking more than once.
  • A known transition should trigger proceed, diagnose, escalate, or timeout behavior.

Do not use when

  • The harness already tracks the background agent, task, or workflow and will wake you.
  • One immediate status check is enough.
  • No absolute deadline or maximum poll count can be set.

output contract

What must be true when it stops.

Produces

  • External-signal and transition definition
  • Cadence, deadline, and poll-count record
  • Observed transition and action taken
  • Final state and total wait report

Guarantees

  • Harness-tracked background work is never polled.
  • Every watch has a durable deadline or maximum poll count.
  • Active cadence stays at or below 270 seconds; idle cadence uses 1200 seconds or more, avoiding roughly 300 seconds.
  • The watcher acts on observed state rather than probability.
  • Terminal failures escalate instead of being polled indefinitely.

Honest failure states

  • timed_out — the deadline or poll cap fired before the target
  • terminal_failure — the signal reached a red state that will not self-resolve
  • wrong_surface — the requested work is harness-tracked or only needs one check

enforcement

Know which claims are executable.

Scripts are linked when this skill owns deterministic machinery. The remaining rules are explicit operating contracts and must not be marketed as hooks.

Procedural contract

This skill composes platform tools and verification evidence; it does not claim a dedicated guard script.

install and invoke

One skill, three honest surfaces.

skills.sh

Available

Universal skill installation, including Claude Code and Codex.

npx skills add https://github.com/ulpi-io/skills-autonomous-engineering --skill watch-and-act

Claude Code plugin

Available

Install the full plugin, then invoke /watch-and-act or let routing select it from context.

Claude installation →

Codex plugin

Work in progress

A work in progress on the codex-native-plugin branch — build it, start a new session, then invoke $autonomous-engineering:watch-and-act.

Codex installation →