Installs into .claude/skills of the current project.
Are you the author of Extract Concepts?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/yogsoth-ai-extract-concepts)
---
name: extract-concepts
description: "Extract candidate concepts from evidence with explicit source provenance and precision/recall posture."
---
# extract-concepts
## Purpose
Extract domain concepts, definitions, and relation candidates from a source while preserving source wording and scope.
## Input contract
```yaml
required: [source, extraction_scope]
optional: [concept_schema, domain_ontology, relation_cues]
constraints: [each concept requires a source span and an uncertainty state]
```
## Procedure
1. Identify repeated technical terms, defined entities, and salient constructs.
2. Capture source definitions, aliases, scope qualifiers, and examples.
3. Link candidate relations only when the source expresses them.
4. Emit a deduplicated concept set with unresolved meanings retained.
If candidate concepts are identified but still bundle multiple meanings or operations, consider `atomize-concept` as the next tactic.
## Output contract
```yaml
produces: [concept_records, definition_map, alias_map, relation_candidates, ambiguity_log]
delta_fields: [findings, evidence_updates, uncertainties, open_questions]
```
## Quality gates
- Concepts are source-linked and distinct from inferred themes.
- Alias merging preserves conflicting definitions.
## Failure and counterexamples
Do not infer a concept solely from keyword frequency or merge terms with different operational meanings.
## Provenance map
- `resolved: extract-concepts`