AgentSkillsCN

using-valence

学习如何高效运用 Valence 知识基底

SKILL.md
--- frontmatter
name: using-valence
description: Learn how to use the Valence knowledge substrate
user_invocable: true

Using Valence

Valence is your personal knowledge substrate. It stores beliefs, tracks conversations, and learns your patterns over time.

Key Concepts

Beliefs: Facts, decisions, preferences stored with confidence levels and provenance. Each belief tracks:

  • •Content (what is believed)
  • •Confidence (how certain, with multiple dimensions)
  • •Domain path (categorization)
  • •Source (where it came from)
  • •Temporal validity (when true)

Sessions: Conversation tracking at multiple scales:

  • •Micro: Individual exchanges (turns)
  • •Meso: Sessions (one conversation)
  • •Macro: Patterns (across sessions)

Patterns: Behavioral patterns that emerge across multiple sessions, like:

  • •Topic recurrence (what you keep coming back to)
  • •Preferences (how you like things done)
  • •Working style (when you're productive, how you approach problems)

Tensions: Contradictions between beliefs that need resolution. Valence detects when beliefs conflict.

Available Skills

  • •/valence:query-knowledge - Search your knowledge base
  • •/valence:capture-insight - Store something important you've learned
  • •/valence:ingest-document - Add a document to the substrate
  • •/valence:review-tensions - Review and resolve contradictions

MCP Tools

You can also use the raw MCP tools directly:

Knowledge (valence-substrate)

  • •belief_query - Search beliefs by content, domain, or entity
  • •belief_create - Store a new belief
  • •belief_supersede - Update a belief while maintaining history
  • •belief_get - Get a belief with full details
  • •entity_get - Get entity details and related beliefs
  • •entity_search - Find entities by name
  • •tension_list - List contradictions
  • •tension_resolve - Resolve a contradiction

Conversations (valence-vkb)

  • •session_start/end/get/list - Manage sessions
  • •exchange_add/list - Record conversation turns
  • •pattern_record/reinforce/list/search - Track patterns
  • •insight_extract/list - Extract insights to KB

Best Practices

  1. •Query first: Before answering questions about past decisions or preferences, query the KB
  2. •Capture insights: When you learn something important about the user, capture it
  3. •Link entities: When creating beliefs, link them to relevant entities (people, tools, projects)
  4. •Review tensions: Periodically check for and resolve contradictions
  5. •Note patterns: When you observe recurring behaviors, record them as patterns

Example Workflows

Learning a Preference

code
User: "I prefer tabs over spaces"
Assistant:
1. Check if there's an existing preference: belief_query("tabs spaces preference")
2. If new, capture: belief_create("User prefers tabs over spaces for indentation",
   confidence={"overall": 0.9}, domain_path=["preferences", "coding"],
   entities=[{"name": "coding style", "type": "concept"}])

Answering from Memory

code
User: "What did we decide about the database?"
Assistant:
1. Query: belief_query("database decision")
2. Review results for relevant beliefs
3. Answer grounded in the beliefs found
4. Cite sources/sessions where relevant

Recording a Pattern

code
After noticing the user often asks about architecture in the morning:
pattern_record(
  type="working_style",
  description="User tends to work on architecture decisions in morning sessions",
  confidence=0.6
)