From hivemind
Connects a mind to the Hive Mind system. Supports three install types — Docker (containerised in this stack), remote (separate host), or bare-metal (systemd service on this host outside Docker). Scaffolds and registers accordingly.
How this skill is triggered — by the user, by Claude, or both
Slash command
/hivemind:add-mindThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
`$ARGUMENTS[0]` = mind name. Ask if missing.
$ARGUMENTS[0] = mind name. Ask if missing.
Check if minds/$ARGUMENTS[0]/ directory exists:
No directory → Scenario A, B, or D. Ask the user:
Directory exists → Scenario C (re-registration). The mind folder is already there but may not be in the broker.
Collect from the user:
name (from argument)gateway_url (default http://hive_mind:8420 for local, ask for remote)model (e.g. sonnet, gpt-oss:20b-32k)harness (e.g. claude_cli_claude, codex_cli_codex)For Scenario B: verify the external gateway is reachable before proceeding:
curl -sf <gateway_url>/broker/minds > /dev/null && echo "Reachable" || echo "UNREACHABLE"
Write minds/<name>/MIND.md:
Scenario A (new local):
---
name: <name>
model: <model>
harness: <harness>
gateway_url: <gateway_url>
---
# <Name>
<Ask the user for a brief identity description to use as the soul seed>
Scenario B (remote):
---
name: <name>
model: <model>
harness: <harness>
gateway_url: <gateway_url>
remote: true
---
Scenario D (bare-metal local):
Ask the user for the port the mind is running on (e.g. 8421). Set gateway_url to http://localhost:<port>.
Confirm the service is up before proceeding:
curl -sf http://localhost:<port>/health && echo "UP" || echo "NOT RUNNING — start the service first"
If not running, stop and tell the user to start the service, then re-run /add-mind.
No MIND.md is written inside the hive_mind project — the mind project lives at its own path on the host. Skip to Step 5.
For Scenario C: MIND.md already exists — skip to Step 3.
List available templates:
ls mind_templates/*.py
Ask the user which template matches their harness + model family. Copy it:
mkdir -p minds/<name>
cp mind_templates/<selected>.py minds/<name>/implementation.py
Replace the MIND_NAME placeholder:
sed -i 's/MIND_NAME/<name>/g' minds/<name>/implementation.py
Create minds/<name>/__init__.py (empty file).
Skip this step for Scenarios B, C, and D.
For Scenario A: check if the MIND.md has a container: block. If it does, run /generate-compose to update docker-compose.yml with the new mind's service definition, then start the container:
docker compose up -d <name>
If the mind does NOT have a container: block, it runs inside the main hive_mind container (subprocess mode) — skip this step.
curl -s -X POST http://localhost:8420/broker/minds \
-H "Content-Type: application/json" \
-d '{"name":"<name>","gateway_url":"<gateway_url>","model":"<model>","harness":"<harness>"}'
If the response contains an error, stop and surface it to the user.
Create a test session:
curl -s -X POST http://localhost:8420/sessions \
-H "Content-Type: application/json" \
-d '{"owner_type":"test","owner_ref":"add-mind-verify","client_ref":"add-mind","mind_id":"<name>"}'
Extract the session ID from the response. Send a test message:
curl -s -X POST http://localhost:8420/sessions/<session_id>/message \
-H "Content-Type: application/json" \
-d '{"content":"Respond with exactly: registration verified."}'
Check if the response stream contains "registration verified". Then clean up:
curl -s -X DELETE http://localhost:8420/sessions/<session_id>
If routability check fails, surface the error clearly but do NOT roll back the registration — the mind is registered, it just couldn't respond yet.
Summarize:
npx claudepluginhub danielstewart77/hivemind-plugin --plugin hivemindProvides 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.