Synthesize cited answers from project Brain artifacts with confidence levels and evidence gaps.
npx skills add https://github.com/reddb-io/red-skills --skill think-reddb-io
Please help me install this Agent Skill. Skill: think Source: https://github.com/reddb-io/red-skills/tree/main/plugins/brain/skills/core/think Command: npx skills add https://github.com/reddb-io/red-skills --skill think-reddb-io
brain_think MCP tool (or red-skills-brain think CLI) to produce a concise answer grounded in captured Brain artifacts, with stable citations to artifact rid/id/kind, score signals, excerpts, and source provenance. - Confidence and Gap Reporting: Relays the returned confidence level (none, low, medium, high) and all missing_evidence items explicitly, never filling gaps with uncited model knowledge. - Brain-vs-Memory Routing: Answers questions about human and project knowledge from Brain while directing operational work facts to /memory:recall. - Use Case: Ask "what does Brain say about our caching strategy?" and receive a synthesized answer with citations, a confidence rating, and a clear statement of any evidence gaps, plus a suggestion to capture missing knowledge via /brain:capture. ## Quick Start Ask the agent to use brain think to answer what the project Brain knows about a specific topic and show the citations and confidence level.High-intent search queries and answers about installing and using this skill.
Call the brain_think MCP tool with your natural-language question, or run red-skills-brain think "<query>" from the CLI. It returns a synthesized answer with citations, a confidence level, and explicit evidence gaps.
brain search returns a raw ranked hit list for browsing individual artifacts, while brain think returns a synthesized, cited answer with confidence and missing evidence. Use think when you want an answer, search when you want to inspect sources yourself.
Confidence is computed by brain_think from ranking signals: high-scoring multi-citation results with source provenance yield high, while a single weak hit without provenance yields low or none. The agent relays it as-is and does not choose it.
No. brain think reasons only over human and project knowledge captured in the Brain store, not operational work facts from the current session. For engineering work facts, use /memory:recall and synthesize both results manually.
When confidence is none or low, or missing_evidence is non-empty, the gaps are stated explicitly instead of being filled with model knowledge. You are prompted to run /brain:capture to store the missing knowledge for future recalls.
buriedsignalsClassifies claims against evidence using a grounding ladder, confidence caps, and contradiction checks.
Untangling-SystemsCapture and analyze learnings from multiple Claude Code sessions.
likesjxStore and recall session context via Muninn MCP with subagents.
kaneyxxSearch the wiki and synthesize cited answers with wikilinks.