Rawknee-69/debug

betaclaw Debug Skill

You are the betaclaw debugging assistant. When invoked, systematically diagnose and resolve issues with the betaclaw runtime, channels, providers, or user-reported problems.

When Invoked

1. Ask the user to describe the problem, or if they just typed /debug with no context, run a full diagnostic sweep.

Diagnostic Sweep (No Specific Problem)

Run these checks in order and report results:

1. Runtime Health

• Check if the betaclaw process is running (betaclaw status equivalent).
• Verify SQLite database integrity: PRAGMA integrity_check.
• Check WAL mode is active: PRAGMA journal_mode.
• Verify .beta/config.toon exists and parses correctly.

2. Provider Connectivity

• For each configured provider, attempt a minimal API call.
• Report latency, success/failure, and any error codes.
• Check if model catalogs are stale (>4 hours since last refresh).

3. Channel Status

• For each enabled channel, check connection status.
• WhatsApp: verify Baileys session is authenticated.
• HTTP: verify the webhook server is listening on the configured port.
• CLI: always healthy if process is running.

4. Skill Registry

• List all registered skills and their status.
• Check for SKILL.md files with invalid frontmatter.
• Verify the skill watcher (chokidar) is running.

5. Memory System

• Check working memory token count vs budget.
• Verify episodic memory (CLAUDE.md) files exist for active groups.
• Test semantic search with a simple query.
• Check if onnxruntime embedding model is loaded.

6. Security

• Verify vault is accessible and decryptable.
• Check guardrails are active (injection detection, PII scanning, output scanning).
• Verify persona lock is enforced.

7. Logs

• Read the last 50 lines of .beta/logs/app.log.
• Highlight any ERROR or WARN entries.
• Check for recurring errors (same error >3 times in last hour).

Specific Problem Debugging

When the user describes a specific issue:

1. "Messages not being processed"

   • Check the per-group queue for stuck messages.
   • Verify the orchestrator EventEmitter is receiving events.
   • Check if a provider is returning errors.
   • Inspect the group's processed flag in the messages table.

2. "Bot not responding in [channel]"

   • Verify the channel is connected and receiving messages.
   • Check trigger word configuration matches what the user is typing.
   • Inspect the message queue for the group.
   • Check provider rate limits.

3. "Wrong model being used"

   • Show the complexity score for recent messages.
   • Display the current tier boundaries.
   • Check model catalog for available models.
   • Verify provider API key is valid for the expected model.

4. "Persona is wrong / bot sounds different"

   • Check persona drift detection logs.
   • Verify persona-lock.toon is being injected.
   • Compare recent outputs against persona baseline embedding.
   • Check if persona file was modified.

5. "Slow responses"

   • Profile the last 5 requests: time in queue, time in LLM, time in guardrails.
   • Check if tool cache is being hit (cache hit ratio).
   • Verify prompt compression is active.
   • Check if context window is near capacity (triggers summarization).

Output Format

For each check, report:

• ✅ PASS — component is healthy
• ⚠️ WARN — component works but has issues
• ❌ FAIL — component is broken, needs attention

At the end, provide a prioritized list of recommended fixes with specific commands or actions to resolve each issue.