Synthesizing Institutional Knowledge

Builds the organizational memory schema your AI agent needs to answer why — capturing decision provenance, causal chains, and event context that embedding-based retrieval permanently discards.

Install
cmdop skills install agensi-synthesizing-institutional-knowledge

Synthesizing Institutional Knowledge is a skill for teams building AI agents that must answer questions about organizational reasoning — not just what a policy says, but why it exists, what it replaced, and what constraints produced it.

The skill addresses a structural gap in how organizations store knowledge. Embedding a document preserves its content but discards the reasoning behind it. The skill introduces a three-type knowledge taxonomy to route information to the correct storage target. Declarative knowledge — facts and current-state policies — goes to a vector RAG store, the only category where embeddings are architecturally sufficient. Episodic knowledge — events, incidents, and decisions with timestamps — goes to a temporal store using a full event schema. Causal knowledge — decision rationale, constraint chains, and alternatives considered — goes to a knowledge graph with explicit predecessor and successor edges, enabling graph traversal over decision chains rather than semantic similarity queries.

The skill provides a complete institutional event schema: a JSON structure capturing actors, affected entities, rationale, alternatives considered, constraints, outcome, and causal links. It also provides an ingestion workflow covering live capture at the moment decisions are made and retroactive extraction from legacy documents such as architecture decision records, post-mortems, and meeting notes, using model-assisted extraction with human review for causal edge validation.

This skill suits engineering, compliance, and strategy teams whose agents need to answer queries like “why do we use X” — queries requiring causal graph traversal, not nearest-embedding retrieval.

Use cases

  • Model organizational decision history so an agent can explain why an architectural choice was made
  • Capture causal chains at the time decisions are made to prevent institutional memory loss
  • Retroactively extract causal edges from existing ADRs, post-mortems, and meeting notes
  • Route knowledge to the correct storage backend — vector store, temporal store, or knowledge graph — based on knowledge type
  • Enable agents to answer compliance or policy questions that require tracing constraint chains back to their origin
  • Build an event schema recording actors, rationale, alternatives, and outcomes for organizational incidents

When to use it

  • The agent must answer why-questions about technology, policy, or architectural decisions
  • The organization needs to preserve decision provenance at capture time rather than reconstruct it later
  • Queries require traversing causal predecessor/successor relationships across decision chains
  • Teams are ingesting legacy documents like ADRs or post-mortems and need to extract causal structure
  • Engineering, compliance, or strategy domains where institutional reasoning compounds over time

When not to use it

  • The use case only requires retrieving factual or current-state content, where standard vector RAG is sufficient
  • No knowledge graph or temporal store infrastructure is available to back the non-declarative knowledge types
  • The team needs a ready-to-run MCP server with callable tools rather than a schema and workflow definition
  • The goal is real-time data retrieval from live systems rather than historical reasoning over recorded decisions