claude

dev-loop

Autonomous multi-agent development loop: tickets, worktree implementers, independent review, batched merge gate, ordered merges.

0 on GitHubby wentbackwardv0.9.0MITupdated 4 days agodevelopment
/

Plugin dev-loop from marketplace claude-dev-loop in wentbackward/claude-dev-loop. Commands read from the repo's manifest — 3 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 (6)

  • dev-loop

    Run the autonomous development loop as the orchestrator - tickets from the spec, up to 3 implementer agents in their own git worktrees, independent code review until approval, one batched full-check gate per train of ready branches, ordered squash merges, then run, verify and close. Use when asked t

  • doctor

    Check that this machine and repository are ready for the dev loop - prerequisites (git, gh login, jq or python3, flock, systemd scopes), the project config and its resolved paths, models per agent, the project's permission rules, the marketplace source and pin in the settings, the current-state docs

  • gate

    Run the batch gate - stack one or more approved branches on the latest base in merge order and run the project's full checks once on the stack, with heavy checks queued machine-wide under a memory cap. Use when branches are reviewed and approved and ready to merge.

  • merge

    Merge a green batch-gate train in order, one squash merge per branch, verifying each replayed tree equals the gated one; opens missing PRs from the notes files. Use right after the gate skill reports TRAIN GREEN.

  • setup

    Bootstrap a project for the dev loop - write CLAUDE.md from the template, docs/DECISIONS.md, .dev-loop.json (check, format and gate commands), the loop's permission rules in .claude/settings.json and missing Makefile targets, asking only for what cannot be inferred. Use when starting the dev loop in

  • status

    Show what the dev loop is doing right now without waking the session that runs it - the batch gate and its current stage (with how long ago its log last grew), the heavy-job slots and what holds them, each loop worktree, the last merge, open tickets, the orchestrator's last step, and a verdict (idle

Subagents (8)

  • architect

    Read-only architecture review from the merge evidence (arch-evidence.sh), the component cards, the current-state docs and the PRD; proposes small refactoring tickets and extraction-ladder steps. Use every architecture.cadence merges (default 20), when a signal is over its threshold, and when the use

  • code-reviewer

    Independent strict review of one branch diff against CLAUDE.md, the spec and the ticket before it is merged. Use on every branch, after the implementer reports green, and again after each round of fixes until APPROVE.

  • doc-auditor

    Read-only audit of the current-state docs and component cards against the code and the running system; opens one ticket per drift, with the evidence, and lists housekeeping as a maintenance list in its report. Use every docs.audit_every merges (default 10), when the user asks, and before a release.

  • implementer

    Implements exactly one ticket, test first, in its own git worktree, then pushes the branch and writes the PR notes file. Never opens or merges PRs. Use one per ticket, at most 3 at a time, launched by the dev-loop orchestrator.

  • scenario-judge

    Judges one holdout scenario against the running system (CLI, HTTP or browser) and prints one JSON verdict with how satisfied a user would be and the behaviour that was missing. Run by the gate's scenario stage (scenario-stage.sh), one scenario at a time. Read-only; never edits files.

  • scenario-writer

    Writes or updates the holdout scenarios for one ticket in the scenario store (scenarios.dir), from the spec and the ticket, never from the code. Use when the orchestrator creates a ticket and the project configures scenarios. Never implements and never edits the repository.

  • test-engineer

    Designs and writes the failing tests for one ticket before it is implemented, from its acceptance criteria. Use at the start of a ticket, in the ticket's worktree, when the orchestrator wants test design separate from implementation.

  • verifier

    Validates the outputs of a real run (a job, an import, a backfill, a deploy, a migration) against the spec and independent evidence. Use after every real run of merged work and before closing the ticket that asked for it. Read-only.

Hooks (3)

  • PreToolUse

    "${CLAUDE_PLUGIN_ROOT}/hooks/guard-role.sh"

  • PostToolUse

    "${CLAUDE_PLUGIN_ROOT}/hooks/format-on-edit.sh"

  • Stop

    "${CLAUDE_PLUGIN_ROOT}/hooks/stop-gate.sh"

View source on GitHub

Similar plugins

Matt Pocock's agent skills for real engineering (grilling, spec/ticket flows, TDD, code review, domain modelling and more).

39 skills

mattpocock/skills 323,108

superpowers

claudecodex

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

15 skills · hooks

obra/superpowers 316,836

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

mem0

claudecodex

Cross-session memory and token savings for coding agents.

6 skills · 1 agent · 1 MCP server · hooks

mem0ai/mem0 68,893

context7

claudecodex

Up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your coding agent's context.

1 MCP server

upstash/context7 64,709

Build powerful, stateful CLI interfaces for any GUI application using the cli-anything harness methodology.

5 commands

HKUDS/CLI-Anything 58,665

Maintain dev-loop?

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 pluginslisted