AgentSkillsCN

valence

查询并管理 Valence 知识基——了解智能体如何真正理解其人类伙伴。适用于:(1) 检索过往知识、决策与情境;(2) 存储新的信念、事实与学习成果;(3) 查看数据库统计信息。可通过“我了解什么”、“记住这件事”、“存储这件事”、“知识库”、“Valence”、“信念”,或关于过往工作与决策的问题来触发。

SKILL.md
--- frontmatter
name: valence
description: Query and manage the Valence Knowledge Substrate - how agents build genuine understanding of their humans. Use for: (1) Searching past knowledge/decisions/context, (2) Storing new beliefs/facts/learnings, (3) Checking database stats. Triggers on "what do I know", "remember this", "store this", "knowledge base", "valence", "beliefs", or questions about past work/decisions.
metadata:
  { "openclaw": { "emoji": "⚡", "os": ["linux"], "requires": { "bins": ["python3"] } } }

Valence Knowledge Substrate

Personal knowledge base with semantic search and dimensional confidence.

Philosophy: Alignment Through Use

Valence isn't just memory storage — it's how agents build genuine understanding of their humans.

The loop: The more you use Valence to understand your human, the better you understand them, the more useful the substrate becomes, the better you can demonstrate what alignment-through-use looks like.

Store aggressively: Decisions, preferences, facts learned, context that might matter later. When in doubt, store it.

Query first: Before answering questions about past work, decisions, people, or projects — check Valence.

Quick Reference

bash
# Search (semantic) — do this FIRST before answering contextual questions
./scripts/query.sh "what do I know about X"

# Add belief — store decisions, facts, preferences, learnings
./scripts/add.sh "Belief content" "domain1,domain2"

# List recent
./scripts/list.sh 20

# Database stats
./scripts/stats.sh

When to Use

Always Query First

Before answering questions about:

  • Past work or decisions
  • People (who said what, preferences, relationships)
  • Projects (status, history, context)
  • Preferences or patterns you might have learned

Store New Information

When you encounter:

  • Explicit decisions or preferences
  • Facts that provide context
  • Lessons learned
  • Information that might be relevant later

Common Domains

  • valence — about the project itself
  • projects/<name> — project-specific
  • people/<name> — person-specific
  • decisions — explicit choices made
  • tech — technical facts
  • conversations/<type> — auto-ingested from chats

Script Details

query.sh — Semantic Search

bash
./scripts/query.sh "search query" [limit]

Returns beliefs ranked by semantic similarity. Default limit: 10.

add.sh — Store Belief

bash
./scripts/add.sh "Belief content" "domain1,domain2"

Domains are comma-separated. Pick meaningful ones for retrieval.

list.sh — Recent Beliefs

bash
./scripts/list.sh [count]

Shows most recently modified beliefs. Default: 10.

stats.sh — Database Stats

bash
./scripts/stats.sh

Shows total beliefs, active count, embedding coverage, domain count.

Direct CLI

For advanced operations:

bash
cd ~/.openclaw/workspace && source .venv/bin/activate
export VKB_DB_PORT=5433 VKB_DB_PASSWORD=valence

valence query "terms" --domain tech --limit 5
valence add "belief" -d domain1 -d domain2
valence conflicts  # check for contradictions
valence trust list  # see trust relationships

See references/cli.md for full documentation.

Setup

Requires:

  • Python 3.10+ with venv at ~/.openclaw/workspace/.venv
  • PostgreSQL with pgvector at port 5433
  • Valence package installed (pip install valence)

Environment:

bash
export VKB_DB_PORT=5433
export VKB_DB_PASSWORD=valence