faff-beep-boop

Orchestrates unattended end-to-end backlog delivery runs with budget ceilings and derailment interrupts.

3|Updated Apr 16, 2026
One-click install
npx skills add https://github.com/shftwst/faff --skill faff-beep-boop-shftwst
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: faff-beep-boop
Source: https://github.com/shftwst/faff/tree/main/plugin/skills/faff-beep-boop
Command: npx skills add https://github.com/shftwst/faff --skill faff-beep-boop-shftwst

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running an AI delivery pipeline overnight or fire-and-forget requires an orchestrator that drains prep and build queues without human prompts, enforces spend limits, and parks ambiguous work instead of guessing. This Skill coordinates the full faff pipeline autonomously so a backlog can be processed while nobody is watching. ## Core Features & Use Cases - Full unattended pipeline: Runs tidy, prep queue drain, and build queue drain end-to-end, delegating to existing faff sub-skills in autonomous mode and parking anything ambiguous for later review. - Budget envelopes and sentry interrupts: Enforces wall-clock, build-attempt, token, and dollar-cost ceilings via faff budget check, plus a derailment kill-switch via faff sentry check at every between-units checkpoint. - Phase selection and convergence: Supports --phases build | prep | all for multi-machine coordination and within-run convergence that folds execution-discovered scope into the same run. - Use Case: Point it at a tracker backlog before leaving for the night; in the morning, review the run summary in .faff/runs/ showing shipped issues, open PRs, parked items, and any budget or sentry stops. ## Quick Start Ask the agent to run the backlog unattended overnight with a budget, for example: run beep-boop with --until 06:00 and a cost ceiling so it drains the prep and build queues and parks anything ambiguous.

Frequently Asked Questions about faff-beep-boop

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run an unattended backlog build overnight?▼

Invoke the beep-boop skill with no arguments to run the full pipeline: tidy, prep queue drain, then build queue drain. It runs non-interactively, parks ambiguous issues, and writes a run summary to .faff/runs/ for morning review.

How do I cap spending on an autonomous agent run?▼

Set a BudgetEnvelope with ceilings for wall-clock time (--until), build attempts (--max), tokens, or dollar cost in the .faffrc budget block. The run calls faff budget check at every between-units checkpoint and stops, narrows, or escalates when a ceiling is breached.

Can multiple machines work the same backlog at once?▼

Yes, use --phases to run one primary with all phases plus build-only workers on other machines. Builds are deduplicated by the tracker In Progress claim, and only the primary preps, so no coordination protocol is needed.

What happens when an autonomous run goes wrong?▼

The sentry check runs at every checkpoint and detects thrash, no-progress, repeated failures, and side-effect escapes. Unattended runs act on abort and pause verdicts by stopping dispatch or parking implicated issues, and aborted runs are resumable.

What are the limitations of unattended L3 runs?▼

There is no mid-issue cancellation, so in-flight builds finish even after a budget fires. A crashed build worker can leave a stale In Progress claim that no peer reclaims until a reclaimer ships, requiring manual cleanup.