extract-entities-and-claims

name: extract-entities-and-claims description: Extracts candidate entities, concepts, claims, and chronology from a policy-cleared evidence package without opening a write path. Use when a governed synthesis workflow needs a deterministic extraction bundle before drafting or escalation.

Extract Entities and Claims

Overview

Use this skill to turn a policy-cleared evidence package into a compact extraction bundle that downstream synthesis scripts consume deterministically. The skill has an executable logic directory (logic/extract_entities.py) that is invoked by the CI-3 PR Producer workflow. It calls GitHub Models API to extract named entities and concepts, writes the bundle to a caller-specified temp path, and returns control to the caller without any wiki write.

Classification

Mode: read-only only
MVP status: Active
Execution boundary: Extraction and bundling only. Output bundle is written to a caller-supplied temp path; no governed wiki write is opened by this skill.

When to Use

policy-arbiter has cleared a source-backed package for controlled drafting
synthesis-curator needs a normalized claim and entity inventory before deciding on create-versus-update planning
A reviewer needs chronology, concepts, and candidate entities separated from raw package prose
The package contains ambiguity that must be surfaced as open questions instead of being guessed away
Drafting should stay deterministic and scoped before any durable follow-up

Contract

Input: a policy-cleared package, cited evidence set, and any scope memo from policy-arbiter
Output: an extraction bundle naming candidate entities, concepts, claim units, chronology notes, cited support, and unresolved gaps
Handoff artifact: an extraction bundle containing candidate page targets, SourceRef-backed claim inventory, chronology notes, and open-question flags
Escalation artifact: an extraction ambiguity note describing unresolved identity, claim-boundary, or source-scope conflicts
Handoff rule: extraction results go to synthesis-curator or back to knowledgebase-orchestrator for governance follow-up; they do not become direct page writes

Assertions

Only policy-cleared, evidence-backed inputs may be processed
Claims stay attributable to cited evidence rather than blended into unsupported narrative
Extraction remains separate from page identity, publication, or persistence decisions
Ambiguity is preserved as an open question or escalation, not collapsed
No direct durable write path opens from this skill

Procedure

Step 1: Confirm governed scope

Read the policy-arbiter scope memo and the cited evidence package so the extraction stays inside the cleared boundary.

Step 2: Inventory entities, concepts, and claims

List candidate entities, concepts, claim units, and relevant chronology markers with their cited support.

Step 3: Separate structure from draft prose

Record what appears supported without deciding final page wording, merge behavior, or publication status.

Step 4: Capture unresolved questions

Flag identity ambiguity, unsupported comparisons, or claim-boundary conflicts so record-open-questions or Human Steward review can pick them up deterministically.

Step 5: Route the bundle

Hand the extraction bundle to synthesis-curator for controlled drafting, or return to knowledgebase-orchestrator when scope, evidence, or policy status is no longer sufficient.

Boundaries

Do not read unadmitted raw/inbox/** material directly from this skill
Do not decide final page identity outside the named ontology and taxonomy contracts
Do not convert extraction notes into a direct write under wiki/
Do not bypass knowledgebase-orchestrator, evidence-verifier, or policy-arbiter when durable follow-up is requested

Verification

Input package is explicitly policy-cleared and scope-bound
Entities, concepts, claims, and chronology are recorded with citations
Ambiguity is preserved in handoff or escalation artifacts
The result routes to synthesis-curator or back through governance
No direct write or persistence side effect is introduced

name: extract-entities-and-claims description: Extracts candidate entities, concepts, claims, and chronology from a policy-cleared evidence package without opening a write path. Use when a governed synthesis workflow needs a deterministic extraction bundle before drafting or escalation.