claude

acf-whip

The ACF cycle, out of the box: /whip is the front door (where you are, and the one command to run next). Small work goes straight to /task and /go; risky work gets a plan in docs/cycles/ that you approve once, bound to its content hash, after which /task --plan decomposes it, /go builds it, /whip shows per-deliverable coverage, and /retro closes the cycle with one-rule lessons and an action plan that feed the next plan. /whip --auto runs the whole loop hands-off until a fresh judge agent finds the goal met; plans still wait for your yes unless you opt in with --approve. Spec reviewers and workbenches (GapTrap ships as the example) plug in with one committed file. Workflow-native backlog automation. /task decomposes a scope into atomic, dependency-aware tasks, each with a role; /go runs each unit as its OWN nested workflow — the shipped harness (implement, quality gates, an adversarial review) run by the generic whippet worker, which loads the role's skill on demand (an installed specialist agent is still honoured) — in isolated git worktrees (one per slug), reviews each accepted diff (once, when a configured gate passed and the harness's own reviewer accepted that commit) and integrates via LP-CAS, looping until every task is accepted or blocked. A bare /go runs your FOCUS: the tasks /task just created, or the ones you selected with /whip-focus — so the ordinary path needs no arguments and still runs only your own work, while /go --all drains the whole backlog and /go <ids|MUL-1234|a natural-language filter> runs exactly what you name. On a shared board a bare /go with no focus refuses rather than draining everyone's queue. The snapshot is narrowed in the STORE (--only emits just the closure of the named tasks) and carries a relay header the orchestrator verifies, so a large board can no longer be silently truncated on its way through a scan agent and reported back as missing task ids. An integrity FATAL now holds out the ONE task it names instead of stopping the ru

2 on GitHubby ACF Tools Teamv3.7.1Apache-2.0updated 9 days agoproductivity
/

Plugin acf-whip from marketplace acf-tools in fib7/acf-tools. Commands read from the repo's manifest — 9 days ago.

This plugin runs code on your machine — its hooks fire automatically on events like tool use. The exact commands are listed below; read them before installing.

What it installs

Skills (13)

  • clean-checkout-before-task

    Mandatory first step for any agent that mutates code. Resolves the task's Worktree slug, lazy-creates an isolated git worktree on a derived branch, validates the branch, hydrates the dev env from .worktreeinclude, stashes unrelated changes, and logs the clean-checkout token. Gives each task its own

  • go

    Continuous backlog execution until 100% of tasks reach a terminal state (accepted/blocked)

  • integrate-lpcas

    LP-CAS integrator. Lands one accepted task's single-parent candidate on the live trunk by cherry-picking it, co-committing an in-tree ledger marker, gating the union tree, and CAS-updating trunk. A task counts as integrated when its marker is in the live-trunk tree, never because of a markdown field

  • research

    Research a topic with MCP tools (Perplexity, Context7, Figma), write a temporal (YYYYMMDD-dated) research document, and generate a backlog task from the findings

  • retro

    Close a planned acf-whip cycle, and learn from it when something went wrong. A clean cycle closes in one script call with zero agents. When a trigger fired (a task reworked, a task stuck, a deliverable not met), a retro agent writes the frozen analysis, at most 3 one-rule lessons, and a retro-action

  • task

    Decompose a scope into atomic tasks using task-engineer agents. It only orchestrates and never implements anything.

  • whip-add-agents

    Replaced by /whip-add-skills in acf-whip 3.7.0. Writes nothing; it only tells you which command to run instead.

  • whip-add-skills

    Role skill + manifest generator. Detects the repository's stack and writes each missing role skill (.claude/skills/{role}/SKILL.md, which /go runs on the whippet worker) with its Priority 1/2/3 manifest and quality-gate docs, following ACF conventions. Never writes agent files. --migrate converts th

  • whip-doctor

    Backlog integrity check for ACF Whip, on every task-store backend (files, beads, multica, linear). Scans the task store and the git worktree registry for ~20 invariants (status/folder mismatch, collapsed task files, dangling or cyclic dependencies, missing/duplicate worktree slugs, orphaned worktree

  • whip-focus

    Manage the run focus, the small per-checkout list of tasks a bare `/go` runs. Set it to what you are working on now, add to it, drop from it, or clear it. It is never committed and never shared, so two people on one board have two focuses. Use when a bare `/go` should run your own tasks rather than

  • whip-init

    Initialize ACF Whip backlog workflow in a project (task store, quality gates, agents, role skills, CLAUDE.md sections)

  • whip-move

    Move one acf-whip backlog task to another state without hand-editing it. `draft` pauses a task and `archive` cancels it; the two are the supported way to park work on every backend. It can also set Integrated-At (files store only) or add a Dependency. Resolves the task by stem or id and runs the val

  • whip

    The front door of the acf-whip cycle (PLAN, DECOMPOSE, BUILD, VERIFY, RETRO). Bare /whip says where you are and prints the one command to run next. /whip \"<goal>\" sends small work straight to /task and risky work to a plan you approve with one yes, which then starts /task --plan itself. /whip <fil

Hooks (3)

  • SessionStart

    "${CLAUDE_PLUGIN_ROOT}/hooks/handlers/session-start.py"

  • PreToolUse

    "${CLAUDE_PLUGIN_ROOT}/hooks/handlers/publish-guard.sh"

  • Stop

    "${CLAUDE_PLUGIN_ROOT}/hooks/handlers/autopilot-stop.sh"

View source on GitHub

More from fib7/acf-tools

All 7 plugins →

Convention enforcement mode - The stern librarian enforces ACF naming and organization standards through context injection. Auto-initializes docs/acf/organization.md, audits compliance, and ensures everything stays in its proper place.

2 skills · hooks

2

Two skills for second opinions: `advisor` consults another LLM (thinking on by default) for one focused take; `board` convenes a multi-persona panel hypertailored to the decision, run as parallel advisor.py calls and synthesized by the orchestrator with weighting on the user-chosen priority dimension. Calls any OpenAI-compatible endpoint; comma-delimited ADVISOR_MODEL declares a multi-model pool for round-robin or cross-model triangulation. Per-family param-shape routing (Claude adaptive, Gemini thinking_level, OpenAI reasoning_effort) with ADVISOR_PROFILE override and shape-swap retry. Falls back to a local-team mode (Claude subagents roleplay personas) when no endpoint is configured.

2 skills

2

Production verification mode - Ruthless honesty enforcer with zero tolerance for false production claims. Binary outcomes only. Perfect for deployment readiness checks and status reporting.

hooks

2

Reasoning validation mode - Intellectually aggressive challenger who attacks every gap in logic and consistency. Perfect for requirements clarification, architecture review, and agent output validation.

hooks

2

Status line HUD: context trend toward the auto-compact point, self-priced session cost (incl. subagents), prompt-cache countdown with rebuild cost, rate limits with a pace projection, PR, acf-whip run status, subagent rows; fits the terminal; configured in /config; personal settings only; self-upgrading

hooks

2

acf-nitro

claude

Parallel execution enforcement mode - Enthusiastic speed demon who catches sequential execution violations and calculates time waste. Enforces independent tool calls in one message, and subagents only for independent work that needs a fresh context.

hooks

2

Similar plugins

superpowers

claudecodex

Core skills library for Claude Code: TDD, debugging, collaboration patterns, and proven techniques

15 skills · hooks

obra/superpowers 316,836

caveman

claude

Talk like caveman. Cut filler. Keep technical details explicit.

22 skills · 1 command · 3 agents · hooks

JuliusBrussee/caveman 119,462

claude-mem

claudecodex

Persistent memory for Claude Code: the next session starts where the last one left off.

29 skills · 1 MCP server · hooks

thedotmack/claude-mem 116,618

Taste Skill

claudecodex

tasteskill 2.5.1: professional company websites at the figma.com and x.ai level, with one section per screen, real photographs, real product UI, effects kept for tech and gaming pages, and a real-browser checker and finish gate.

1 skill

Leonxlnx/taste-skill 114,391

career-ops

claudecodex

Open-source AI job search that runs locally in your AI coding CLI: scan job portals, evaluate listings into a structured A-H report with a 1-5 score, tailor your CV, track applications

5 skills · hooks

career-ops-hq/career-ops 82,253

mem0

claudecodex

Cross-session memory and token savings for coding agents.

6 skills · 1 agent · 1 MCP server · hooks

mem0ai/mem0 68,893

Maintain acf-whip?

This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.

claude & codex plugins7 plugins