From agent-atelier
Initialize the orchestration workspace with default state files and directory structure. Use when bootstrapping a new project, when orchestration state is missing or corrupt, when the user says 'init', 'initialize', 'bootstrap', 'set up the workspace', 'start a project', 'prepare for agent-atelier', or 'create state files'. Also triggers automatically before /agent-atelier:run if .agent-atelier/ does not exist. Safe to re-run — only creates files that don't already exist.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-atelier:initThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
- First-time project setup
/agent-atelier:run when .agent-atelier/ is missing--root)/agent-atelier:init # auto-detect git root
/agent-atelier:init --root /path/to/repo # explicit root
Detect root. Run git rev-parse --show-toplevel, or use --root <path> if provided. If neither works, exit with code 3.
Create directories. mkdir -p for each (no-ops if they exist):
.agent-atelier/.agent-atelier/human-gates/open/.agent-atelier/human-gates/resolved/.agent-atelier/human-gates/templates/.agent-atelier/attempts/Create state files. For each file, check existence first. Only write if missing. Use the default shapes from references/state-defaults.md (in this plugin's parent references/ directory). Replace every <now> placeholder with the current UTC timestamp (YYYY-MM-DDTHH:MM:SSZ).
| File | Source Template |
|---|---|
.agent-atelier/loop-state.json | Default loop state |
.agent-atelier/work-items.json | Default work items |
.agent-atelier/watchdog-jobs.json | Default watchdog jobs |
.agent-atelier/human-gates/_index.md | Default gate dashboard |
.agent-atelier/human-gates/templates/human-decision-request.json | Gate template |
Check for incomplete transactions. If .agent-atelier/.pending-tx.json exists, a previous state-commit was interrupted. Replay it:
cat .agent-atelier/.pending-tx.json | \
<plugin-root>/scripts/state-commit --root <repo-root> --replay
Where <plugin-root> is the resolved path to plugins/agent-atelier (e.g., via ${CLAUDE_SKILL_DIR}/../..). The replay is idempotent: already-written files are skipped. Report the recovery in output.
Report results. Print a summary:
Initialized orchestration workspace at /path/to/repo
Created: loop-state.json, work-items.json, watchdog-jobs.json
Already existed: human-gates/_index.md
| Code | Meaning |
|---|---|
0 | Success — workspace initialized or already up to date |
1 | Usage error (invalid arguments) |
3 | Not inside a git repository and no --root given |
4 | Runtime failure (disk error, permission denied, WAL replay failure) |
Returns JSON to stdout:
{
"request_id": "<id>",
"accepted": true,
"committed_revision": 1,
"changed": true,
"root": "/path/to/repo",
"artifacts": [
".agent-atelier/loop-state.json",
".agent-atelier/work-items.json",
".agent-atelier/watchdog-jobs.json"
],
"wal_recovered": false
}
"changed": false when all files already exist. Diagnostic messages go to stderr.
Re-running when files exist returns "changed": false. WAL replay skips files whose revision already matches the target.
--root: Exit 3. Suggest --root /path/to/repo..pending-tx.json) is preserved for manual inspection.state-commit script not found: Exit 4. Report the expected path (<plugin-root>/scripts/state-commit).Z suffix.npx claudepluginhub ether-moon/agent-atelier --plugin agent-atelierProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.