Skip to content

Fix/wake word interruption#18

Open
beancookie wants to merge 4 commits intomainfrom
fix/wake-word-interruption
Open

Fix/wake word interruption#18
beancookie wants to merge 4 commits intomainfrom
fix/wake-word-interruption

Conversation

@beancookie
Copy link
Copy Markdown
Owner

No description provided.

beancookie and others added 4 commits April 14, 2026 22:37
Remove misleading tool call examples (Tool: xxx / Input: {}) that caused
the LLM to output tools as plain text instead of using the API's tool_calls
mechanism. Skills now focus on "when to use" descriptions rather than
showing text-formatted tool invocations.

Changes:
- mcp-servers: describe tools by purpose, not invocation format
- lua-scripts: remove text-formatted lua_run/lua_eval examples
- skill-creator: remove MCP/Lua content (now separate skills), focus on core skill creation
- weather: removed (obsolete, web_search handles this)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Playing TTS audio logs were spaming at INFO level during normal
operation. Change to DEBUG to reduce log output.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Parser expects `## test_server` as section header. Change from inline
format `**test_server**: ...` back to proper markdown section.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Previously tools_json was fetched once at agent startup, but MCP tools
like mcp_server.tools_list are only registered after mcp_connect succeeds.
This caused the LLM to not see MCP tools and output them as plain text.

Now re-fetches tool_registry_get_tools_json() before each LLM call so that
dynamically registered tools (MCP, etc.) are available immediately.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant