AgentSkillsCN

capture-insight

将重要洞见或决策记录到知识库中

SKILL.md
--- frontmatter
name: capture-insight
description: Capture an important insight or decision to the knowledge base
user_invocable: true
args:
  - name: content
    description: The insight to capture
    required: true
  - name: domain
    description: Domain classification (e.g., "tech/architecture", "preferences")
    required: false
  - name: confidence
    description: Confidence level 0-1 (default 0.8)
    required: false

Capture Insight

Store an important piece of information in the Valence knowledge base.

Instructions

  1. •First, search for similar existing beliefs using mcp__valence_substrate__belief_query
  2. •If a very similar belief exists (>90% match), consider:
    • •Using mcp__valence_substrate__belief_supersede to update it
    • •Boosting its confidence through corroboration
    • •Noting any differences as a potential tension
  3. •If this is new information, use mcp__valence_substrate__belief_create with:
    • •The content as provided
    • •Appropriate confidence level (default 0.8 for user-stated facts)
    • •Domain path parsed from the domain argument (split on "/")
    • •Current session as source (use VALENCE_SESSION_ID if available)
  4. •Extract and link relevant entities (people, tools, projects, concepts)
  5. •Confirm what was captured to the user

Content to Capture

{{ content }}

{% if domain %}

Domain

{{ domain }} {% endif %}

{% if confidence %}

Confidence

{{ confidence }} {% endif %}

Execution

  1. •Search for duplicates first
  2. •Parse domain path (e.g., "tech/architecture" -> ["tech", "architecture"])
  3. •Identify entities in the content (look for proper nouns, tool names, etc.)
  4. •Create the belief with appropriate metadata
  5. •Report what was stored and any entities linked