Claude-skill-registry hive
Orchestrate a multi-step workflow as coordinated sub-tasks; use when a task is large and benefits from structured coordination.
install
source · Clone the upstream repo
git clone https://github.com/majiayu000/claude-skill-registry
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/hive" ~/.claude/skills/majiayu000-claude-skill-registry-hive && rm -rf "$T"
manifest:
skills/data/hive/SKILL.mdsource content
Hive
Overview
Use mprocs to run a queen plus workers in parallel with shared session state.
Inputs
- Session name
- Worker count (default 4, max 5)
- Task description
Workflow
- Verify prerequisites:
git --versionmprocs --version
- Parse session name and worker count.
- Choose
:GEMINI_MODEL
for research and discoverygemini-3-flash-preview
for code generation or implementationgemini-3-pro-preview
- Run a lightweight pre-scan with Codex CLI (commands below).
- Create
and write.hive/sessions/<session-id>
.tasks.json - Write
andqueen-prompt.md
prompts.worker-*.md - Write
and launch mprocs..hive/mprocs.yaml
Pre-scan Commands
codex exec -m gpt-5.2 -s read-only -c model_reasoning_effort="low" --skip-git-repo-check \ "Scan this codebase and identify key entry points and core logic for: {TASK_DESC}. Return file paths with brief notes." codex exec -m gpt-5.2 -s read-only -c model_reasoning_effort="low" --skip-git-repo-check \ "Find high-coupling files, config files, and package definitions for: {TASK_DESC}. Return file paths with brief notes."
tasks.json Template
{ "session": "{SESSION_ID}", "created": "{ISO_TIMESTAMP}", "status": "active", "session_status": "active", "task_description": "{TASK_DESC}", "workers": {WORKER_COUNT}, "tasks": [], "synthesis": { "status": "pending", "result_file": "results.md" } }
mprocs.yaml Template
procs: queen: cmd: ["claude", "--model", "opus", "--dangerously-skip-permissions", "Read .hive/sessions/{SESSION_ID}/queen-prompt.md"] cwd: "{PROJECT_ROOT_FORWARD}" worker-1: cmd: ["claude", "--model", "opus", "--dangerously-skip-permissions", "Read .hive/sessions/{SESSION_ID}/worker-1-prompt.md"] cwd: "{PROJECT_ROOT_FORWARD}" worker-2: cmd: ["powershell", "-NoProfile", "-Command", "gemini -m {GEMINI_MODEL} -y -i 'Read .hive/sessions/{SESSION_ID}/worker-2-prompt.md'"] cwd: "{PROJECT_ROOT_FORWARD}" worker-3: cmd: ["powershell", "-NoProfile", "-Command", "codex --dangerously-bypass-approvals-and-sandbox -m gpt-5.2 'Read .hive/sessions/{SESSION_ID}/worker-3-prompt.md'"] cwd: "{PROJECT_ROOT_FORWARD}"
Launch
mprocs --config .hive/mprocs.yaml
Output
- Session folder with prompts, tasks, and results
- Final synthesis in
.hive/sessions/<session-id>/results.md