claudecodex

atelier

Delegate work to model-tiered agents instead of doing it all in one session: size the task, pick a delegation architecture, and dispatch scoped slices to scouts, builders, reviewers, and managers. Skills cover orchestration, session handoff, backlog waves, and a review cycle that runs before work lands. Hooks watch the context watermark, keep the handoff fresh and surface it on cold start, isolate writing workers in their own git worktrees, hold protected config in custody, and refuse a worker git call that would sweep away a sibling's uncommitted work or land on a protected branch. Enforcement follows the main checkout into a linked worktree, so an isolated worker is no longer an unguarded one. Run /atelier:activate to arm it in a project and see what each hook resolved. Release history: https://github.com/hsb3/dotfiles-agents/releases.

0 on GitHubby Henry S. Burden IIIv0.39.9MITupdated 29 days ago
/

Plugin atelier from marketplace dotfiles-agents in hsb3/dotfiles-agents. Commands read from the repo's manifest — 28 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 (9)

  • activation

    Create and verify the harness-appropriate per-project activation file that arms atelier's hooks. Use when someone asks to turn on, configure, or check atelier enforcement (custody, worker context, worktree isolation, handoff routing) in a project, or when a hook that should be firing appears silent.

  • comment-hygiene

    Strip history and commentary out of source comments before work lands — harvest the reasoning onto its tracker item, keep only what a reader would break something without. Use before a PR or a milestone commit, when a file's comments read as narrative, or when invoked by layer-cycle's refine phase.

  • compact-handoff

    Update and verify the session handoff, then prepare a KEEP/DROP compaction prompt that preserves active agents and dependent commands. Use when asked to compact with a handoff or prepare session compaction.

  • delegation

    How to split a substantial task across agents: the three-layer model, the architecture matrix, effort and model-tier calibration, brief rules, and the never-delegated floor. Use when sizing a feature build, refactor, migration, audit, or multi-file fix; when the work arrived as a goal ("clean this u

  • deletion-pass

    Simplify a module to irreducible against its contract — remove or collapse every line that cannot name the commitment it keeps, without changing observable behavior. Use when asked to simplify or tighten a module against its spec, to produce a dry-run deletion report, or when invoked by layer-cycle

  • handoff

    Maintain the project's session-handoff so a brand-new session can pick up work cold - a file this skill writes directly (_meta/HANDOFF.md, HANDOFF.md, .claude/HANDOFF.md, or a per-project override), or an external tracker/board it signals via a freshness stamp when the project's handoff key is set t

  • layer-cycle

    Drive a module through create → evaluate → refine cycles until convergence or budget exhaustion, translating findings into scoped fix briefs. Use when asked to "run the cycle", take a module through review and refinement, or iterate a module against a contract. Companions — rubric-panel (evaluate),

  • rubric-panel

    Score one or more code artifacts against an anchored rubric using a persona-diverse judge panel; produces dimension scores plus findings classified as defect, noise, spec-hole, or undeclared-commitment. Use when asked to "score this implementation", "judge these two solutions", "rate this module aga

  • waves

    Drive a project's open backlog to closed through triage, wave planning, and isolated agent crews — the full loop: refresh the triage view, group buildable items into branch-sized waves, launch worktree-isolated teams via the delegation skill, verify and land each wave in order, reconcile, and extern

Slash commands (2)

  • /atelier:activate

    Arm atelier in this project — create the selected per-project activation file if it is missing, then report per key what the hooks actually resolved, including any key that is present but silently doing nothing.

  • /atelier:prepare-compact

    Update the handoff and prepare a compaction prompt that preserves live work.

Subagents (5)

  • builder

    Scoped implementation inside an owned file list against explicit acceptance criteria. Defaults to the mid tier for well-specified bounded edits; dispatch one tier up (on this harness, model:opus) when the slice has coupled logic or being wrong is expensive to unwind.

  • code-reviewer

    Senior-engineer code-quality review focused on simplification, minimalism, and elegance. Flags over-engineering, unnecessary abstraction, and needless complexity, then shows the cleaner form with before/after. Use after writing or modifying code for a quality and simplification pass, distinct from t

  • manager

    The management layer between strategy and execution — owns a wave or a coupled chain end to end: turns the definition of done into worker briefs, spawns and sequences its own scouts, builders, and reviewers, verifies their output, and reports one proof package upward. The default for non-trivial wor

  • reviewer

    Adversarial, report-only verification — re-derives each claim from its cited source and re-runs its commands; never edits, fixes, or writes into the repo under review. Use whenever a claim or diff will drive further changes; a producer self-report is a hypothesis, not proof.

  • scout

    Read-only recon — locate definitions, confirm presence/absence, inventory a scope, or reconcile evidence across files; returns a conclusion with path:line evidence, never a file dump. Runs no commands unless the brief names read-only ones. Defaults to the light tier; dispatch one tier up (on this ha

Hooks (19)

  • PostToolUse

    DELEGATION_WATERMARK_SOFT="${DELEGATION_WATERMARK_SOFT:-25}" DELEGATION_WATERMARK_REFIRE_EVERY="${DELEGATION_WATERMARK_REFIRE_EVERY:-15}" DELEGATION_WATERMARK_FLOOR_COMMANDS="${DELEGATION_WATERMARK_FL

  • PostToolUse

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/context-watermark/hook.py"

  • PreCompact

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/handoff-freshness-guard/hook.py"

  • PreToolUse

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/handoff-freshness-guard/hook.py"

  • PreToolUse

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/comment-hygiene-gate/hook.py"

  • PreToolUse

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/live-worker-git-guard/hook.py"

  • PreToolUse

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/worker-git-scope-guard/hook.py"

  • PreToolUse

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/skills-cli-guard/hook.py"

  • PreToolUse

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/zsh-modifier-lint/hook.py"

  • PreToolUse

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/config-custody/hook.py"

  • PreToolUse

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/worktree-isolation/hook.py"

  • SessionStart

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/session-handoff-surfacer/hook.py"

  • SessionStart

    LANE_SNAPSHOT_INTERVAL="${LANE_SNAPSHOT_INTERVAL:-180}" python3 "${CLAUDE_PLUGIN_ROOT}/hooks/lane-snapshot/hook.py"

  • SessionStart

    BRANCH_ACTIVITY_PEER_TTL_SECONDS="${BRANCH_ACTIVITY_PEER_TTL_SECONDS:-3600}" BRANCH_ACTIVITY_MAX_BYTES="${BRANCH_ACTIVITY_MAX_BYTES:-1048576}" BRANCH_ACTIVITY_LOG_COMMITS="${BRANCH_ACTIVITY_LOG_COMMIT

  • Stop

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/subagent-telemetry/hook.py"

  • SubagentStart

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/worker-context/hook.py"

  • SubagentStop

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/subagent-telemetry/hook.py"

  • SubagentStop

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/manager-package-gate/hook.py"

  • UserPromptSubmit

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/context-watermark/hook.py"

View source on GitHub

More from hsb3/dotfiles-agents

All 12 plugins →

The home for every skill that stands on its own and has no topical plugin of its own, in one install instead of a shelf of one-skill plugins: harness and environment config, test quality, research and decision tooling, chart design, API and TUI craft, repo and project setup, and sandboxed opencode instances. Membership is derived, not curated — a gate re-reads the skill bodies every run and keeps out anything that needs a companion agent or hook, anything a topical plugin already owns, plus a documented exemption class for skills that prescribe an opt-in system the consumer chooses deliberately, in-repo or external. Release history: https://github.com/hsb3/dotfiles-agents/releases.

14 skills

0

code-desk

claudecodex

Set a project's starting conditions and run the work through them: interview the owner into a written quality contract, build and red-prove the one gate command that enforces it, then keep the release loop honest — the review findings a green check hides, recurring status comms and authored presentations with portable PptxGenJS packages, existing-deck editing and rendered verification, READMEs backed by live screenshots, and the project-memory taxonomy. The in-repo _meta planning system that used to ship here now lives in the mise-en-place plugin. Release history: https://github.com/hsb3/dotfiles-agents/releases.

6 skills · 1 command · 1 agent

0

An in-repo _meta/ planning system: a documented directory standard, a read-only compliance audit against it, an additive-only scaffolder that fills exactly the gaps the audit found, and a planning desk under _meta/plans/ holding one build plan per unit of work. The supporting standards ship with it: the project-memory taxonomy whose checklist rows the audit and the scaffold both read, and the item-body form the desk writes tracker items to, carried as a dependency from its topical home in board-desk. The standard, the audit, and the scaffold assume the repo tree is the system of record, so install them only where that is true. The desk is the exception: it reads work items from a tracker through a pluggable adapter, and a Kata adapter ships. Release history: https://github.com/hsb3/dotfiles-agents/releases.

6 skills

0

pocketbase

claudecodex

Build and operate PocketBase backends: an operational skill for driving a running instance (collection and record CRUD, auth, backups, migration generation, Go hooks and custom routes), a design skill of 63 prioritized rules for schema, API rules, auth flows, query performance, realtime, files, and deployment, and three subject-matter agents that carry the backend laws into a delegated brief: a builder, an adversarial reviewer, and a security auditor, each proving its claims from its own clean-room server. Release history: https://github.com/hsb3/dotfiles-agents/releases.

3 skills · 3 agents

0

diagrams

claude

Structural diagrams for repo docs, decks, and briefings, with consistent SVG and PNG output: Mermaid for GitHub-rendered markdown, Python diagrams for cloud architecture with provider icons, draw.io reading and headless export, Excalidraw sketches, and raw Graphviz. Data charts belong to the dataviz skill, deck theming to presentations. Release history: https://github.com/hsb3/dotfiles-agents/releases.

4 skills

0

Build Obsidian plugins against the real API — lifecycle, settings, vault operations, commands, chat and copilot sidebar UIs, in-plugin MCP servers over Streamable HTTP — and automate vaults from the terminal with the official Obsidian CLI. Release history: https://github.com/hsb3/dotfiles-agents/releases.

4 skills

0

Maintain atelier?

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 plugins12 plugins