muninn

Orchestrate cross-instance Claude Code debugging via claude-hivemind — send test prompts to a Muninn bot peer (jarvis, melosys, capra, …), inspect the resulting trace in the Muninn DB, and coordinate with peer agents (e.g. huginn for search/retrieval changes) to iterate on behavior. Use this whenever you need to test how a Muninn bot reacts to an upstream change, debug search quality across the muninn+huginn split, validate that a tweak landed correctly, or run any multi-instance experiment that has one peer making changes and another driving the test loop. Triggers on phrases like "iterate with peers", "test bot via hivemind", "coordinate with huginn", "send test to melosys", "iterate on search ranking", "verify the change in the bot", or any debugging task that spans muninn (the orchestrator) and another peer instance.

Access and query the Muninn PostgreSQL database running in Docker. Use this skill whenever you need to run SQL queries, inspect data, debug database issues, check table contents, or understand the schema. Triggers on database queries, SQL, checking data, table inspection, "what's in the DB", row counts, data debugging, or any task requiring direct database access in the Muninn project.

Deep knowledge of the Muninn AI assistant's knowledge pipeline — async Haiku calls, memory extraction, goal detection, schedule detection, prompt assembly, embeddings, token tracking, and the unified scheduler. Use when: (1) Understanding how the knowledge system works end-to-end, (2) Improving or debugging any Haiku pipeline (memory, goals, schedules), (3) Modifying prompt-builder context assembly, (4) Working with embeddings or hybrid memory search, (5) Adding new async extraction pipelines, (6) Debugging token tracking or dashboard stats, (7) Understanding the scheduler runner and task execution. Triggers: "knowledge system", "haiku pipeline", "memory extraction", "goal detection", "schedule detection", "prompt builder", "embeddings", "token tracking", "scheduler", "how does memory work", "how does Jarvis learn", "add a new extractor".

Patterns and pitfalls for building Slack bots with @slack/bolt in Socket Mode. Covers the Assistant API, thinking indicators, event deduplication, thread management, and message formatting. Use when: (1) Building or modifying Slack bot handlers (app_mention, app.message, Assistant), (2) Implementing thinking/typing indicators in channels or threads, (3) Debugging duplicate messages or missing responses in Slack, (4) Working with assistant.threads.setStatus in channel threads, (5) Managing thread tracking and follow-up messages, (6) Formatting messages for Slack (mrkdwn vs HTML), (7) Posting, updating, or deleting bot messages. Triggers: "slack", "bolt", "app_mention", "setStatus", "thinking indicator", "slack thread", "slack channel", "slack bot", "slack assistant", "slack formatting".