aget-expand-ontology

name: aget-expand-ontology description: Expand SKOS ontology vocabularies with web-researched, evidence-backed concepts. Identifies thin topics, proposes expansion areas, researches best-practice concepts, creates bridging interlinkage, writes formal SKOS entries, and enriches matched existing entries. version: 2.1.0 status: active spec: SKILL-019 category: Creation ontology: vocabulary_glob: "ontology/ONTOLOGY_*.yaml" validation_command: "test -f tools/validate_ontology.py && python3 tools/validate_ontology.py --format=report --log-history || echo 'Validation skipped (tools/validate_ontology.py not available)'" quality_history: "ontology/QUALITY_HISTORY.yaml" vocabulary_index: "ontology/VOCABULARY-INDEX.md" project_prefix: "ONTO" staging_file: "ontology/EXPANSION_CANDIDATES.yaml"

/aget-expand-ontology

Expand this agent's SKOS vocabularies with new web-researched, evidence-backed concepts targeting thin topics. Merged from professional-core v1.5.0 + framework-AGET v1.0.0 + supervisor additions (SKILL-019 spec).

Purpose

Automate the research-informed ontology expansion workflow:

1. Diagnose: Identify thin topics and KB gaps
2. Propose: Rank expansion areas by gap size and tier importance
3. Research: Web-search for best-practice evidenced concepts
4. Design: Write formal SKOS entries with interlinkage
5. Enrich: Enhance matched existing concepts with new evidence
6. Verify: Validate no orphans, topic density improved

Input

Optional parameters after skill name:

Examples:

• /aget-expand-ontology — Auto-detect thinnest vocabulary, 3 areas, 7+3 concepts each
• /aget-expand-ontology --vocabulary ontology/ONTOLOGY_personal_ai_systems_v1.0.yaml — Target specific vocabulary
• /aget-expand-ontology --from-kb knowledge/REPORT.md — Document-driven expansion
• /aget-expand-ontology --dry-run — Propose expansion without writing
• /aget-expand-ontology --concepts-per-area 5 --bridging-per-area 2 — Smaller expansion
• /aget-expand-ontology --provenance "Marcus 1:1 surfaced missing meta-concepts for team topology" — Expansion with strategic context

Configuration

The ontology config block in frontmatter defines all paths. Agents MUST customize this block for their vocabulary layout:

ontology:
  vocabulary_glob: "ontology/ONTOLOGY_*.yaml"           # Glob for vocabulary files
  validation_command: "test -f tools/validate_ontology.py && python3 tools/validate_ontology.py --format=report --log-history || echo 'Validation skipped (tools/validate_ontology.py not available)'"
  quality_history: "ontology/QUALITY_HISTORY.yaml"       # Expansion audit trail
  vocabulary_index: "ontology/VOCABULARY-INDEX.md"       # Optional index file
  project_prefix: "ONTO"                                # Expansion project ID prefix
  staging_file: "ontology/EXPANSION_CANDIDATES.yaml"     # Deferred concepts staging

Example configurations:

• Standard agent: vocabulary_glob: "ontology/ONTOLOGY_*.yaml"
• Professional-core: vocabulary_glob: "vocabulary/INTELLIGENCE_VOCABULARY_v*.yaml"
• Multi-vocabulary: vocabulary_glob: "knowledge/frameworks/*-vocabulary-skos.yaml"

Execution

Step 0.5: Prerequisite Validation

Before loading state, validate required files exist:

1. Check vocabulary files match vocabulary_glob from config block
2. Check quality_history file exists (or create empty)
3. Check validation command is available

If prerequisites fail, STOP and advise which files are missing.

Step 1: Load Current Ontology State

Read required files to understand current state:

1. List vocabularies matching config vocabulary_glob
2. If --vocabulary specified, filter to that file only
3. Read target vocabulary (smallest by concept count if multiple)
4. Parse concept scheme, clusters/topics, and concept counts
5. Read quality history from config quality_history path

Step 1b: Capture Expansion Provenance

If --provenance was provided, store it for use in Steps 8 and 10.

If --provenance was NOT provided, prompt:

What triggered this expansion? Briefly describe the event, motivation, or audience (e.g., "Marcus 1:1 surfaced missing meta-concepts" or "pre-release vocabulary gap audit"). Press Enter to skip.

If user provides a response, store as provenance. If skipped, provenance is null (no provenance recorded — expansion proceeds without strategic context).

Provenance structure (used in Steps 8 and 10):

provenance:
  event: "{what happened}"
  motivation: "{why it matters}"
  audience: "{who benefits}"

If the user provides a single string, parse it as event and leave motivation and audience as null.

Step 2: Gap Analysis

From the loaded vocabulary, analyze concept distribution:

For each cluster/topic:
  - current_concepts: count concepts in cluster
  - tier: from topic/cluster metadata (if available)
  - priority_score = (5 - tier) * (target_density - current_concepts)
    where target_density = 10 for top-level, 6 for subtopics, 4 for leaf

Then scan KB artifacts for ungrounded concepts:

Search knowledge/, docs/, .aget/evolution/ for terms not yet in vocabulary

Step 3: Propose Expansion Areas

Rank candidate areas by priority_score and select top N (default 3).

For each proposed area, report:

• Area name: Descriptive label
• Target clusters/topics: Which thin areas this strengthens (with before-after projection)
• Target categories: Which existing categories receive new concepts
• KB evidence: Which KB artifacts contain related ungrounded concepts
• Rationale: Why this area is a natural expansion

Present areas to user. If --dry-run or --plan, present and stop here.

Step 4: Web Research (Parallelized)

For each expansion area, launch a parallel web search agent (Task tool, subagent_type: general-purpose):

Search prompt template:

Search the web for well-established, best-practice concepts in {AREA_NAME}
suitable for a {VOCABULARY_DOMAIN} vocabulary. Focus on concepts with
strong evidence grounding (academic papers, industry reports, standards bodies).

Search for these specific areas:
{SPECIFIC_SEARCH_TOPICS}

Vocabulary context for this expansion:
- Target clusters: {TARGET_CLUSTERS_WITH_COUNTS}
- Key existing concepts to connect to: {KEY_EXISTING_CONCEPTS}
- Hierarchy opportunities: {HIERARCHY_NOTES}

For each concept found, provide:
- Canonical term name
- Brief definition (1-2 sentences)
- Key source/evidence (author, publication, year)
- Why it matters for the vocabulary domain
- Which existing concept(s) it would connect to

Return at least {CONCEPTS_PER_AREA + 4} well-evidenced concepts with sources.

If --from-kb <path> is provided, extract concepts from the document instead. If --skip-web-search is passed, use KB-only analysis.

Evidence quality tiers (target: 60%+ at Tier 1-2):

• Tier 1 (Gold): Peer-reviewed paper, W3C/NIST/ISO standard
• Tier 2 (Silver): Major industry research (McKinsey, Gartner, ACM, IEEE)
• Tier 3 (Bronze): Industry report, conference talk, well-sourced blog

Step 5: Concept Selection and Curation

From web research results, select top N concepts per area:

Selection criteria:

Deduplication: Check each candidate against existing vocabulary prefLabels and altLabels.

Split selected concepts into:

• Domain concepts (N per area): Core concepts for the expansion area
• Bridging concepts (M per area): Concepts connecting new area to existing ontology

Step 5a: Counter-Perspective Gate

After concept selection and before SKOS writing, verify balanced perspective.

Gate check:

1. Classify each selected concept as advocacy (capabilities, benefits, best practices) or counter-perspective (limitations, failures, risks, critiques)
2. Calculate ratio: counter_count / total_selected
3. Compare against threshold: 20-30%

Gate outcomes:

Step 5b: Enhance Matched Concepts

During deduplication, concepts matching existing entries are candidates for enhancement rather than silent discard. Skip if --skip-enhance is passed.

For each matched concept:

1. Assess new evidence: Compare source evidence with existing example and sources fields
2. Determine enhancement value:

3. Apply enhancements (append-only):

   • example: Append new evidence after existing text, with date attribution
   • sources: Append new source reference to list
   • definition: Update ONLY if new evidence materially changes the concept's meaning

4. Track enhancements: Record each for the expansion report

Step 5c: Persist Unused Concepts

Valid concepts not selected for this expansion are written to the staging_file from config block for future consideration.

Step 6: Assign Concept IDs

For each new concept:

1. Determine target category/cluster based on semantic fit
2. Find next available ID in that category
3. Assign ID following vocabulary's naming convention
4. Auto-increment project ID from quality_history

Step 7: Write Formal SKOS Definitions

For each concept, write complete SKOS entry:

{ID}:
  prefLabel: "{canonical name}"
  status: active
  altLabel:
    - "{alternative name 1}"
  definition: "{1-2 sentence formal definition}"
  broader: {parent concept or null}
  narrower: [{child concepts}]
  related: [{related concepts - MUST include >=1 existing concept}]
  example: "{concrete example with source attribution and year}"
  sources:
    - "{Primary source with year}"

SKOS relationship rules:

• related: MUST contain at least 1 existing concept (no orphans)
• broader/narrower: Use when clear hierarchical relationship exists
• sources: MUST contain real, verifiable references (NEVER fabricate)

PrefLabel quality rules (PL-01 through PL-05):

Gate: >10% of new concepts failing PL-01 triggers review before proceeding to Step 8.

Step 7a: Reciprocal Link Verification

After writing SKOS entries and before writing to vocabulary, verify reciprocal link integrity.

For each new concept, check every link it declares:

Include reciprocal link updates in the Step 8 vocabulary write.

Step 8: Write to Vocabulary

If NOT --dry-run:

1. Add SKOS entries to appropriate clusters in vocabulary YAML
2. Apply enhancements to matched existing concepts (unless --skip-enhance)
3. Apply reciprocal link updates to existing concepts
4. Update metadata: concept count, extension projects, version history
5. Write provenance (if captured in Step 1b): Add provenance: sub-field to the new version_history entry:

   - version: "X.Y.Z"
     date: "YYYY-MM-DD"
     provenance:
       event: "{from Step 1b}"
       motivation: "{from Step 1b}"
       audience: "{from Step 1b}"
     changes:
       - "..."
   

6. Write provenance to quality_history (if captured): Add provenance: field to the snapshot entry in the config quality_history file

Step 8b: Update Index

If vocabulary_index path exists in config, update it with new concepts and categories.

Step 9: Validate

Run validation command from config block and verify:

• Quality score maintained or improved
• 0 orphan concepts
• All new concepts have sources
• Cluster/topic density improved as projected
• Concept count matches metadata

Step 10: Report Results

## Ontology Expansion Complete

**Vocabulary**: {name} v{before} -> v{after}
**Concepts**: {before} -> {after} (+{delta})
**Project**: {prefix}-{YYYY}-{NNN}
**Provenance**: {event} (or "none captured")

### Areas Expanded

| Area | Concepts | Clusters | Impact |
|------|----------|----------|--------|
| {name} | {N} domain + {M} bridging | {clusters} | {before}->{after} |

### Enhanced Existing Concepts

| ID | prefLabel | Fields Updated | New Evidence Summary |
|----|-----------|----------------|---------------------|
| {id} | {label} | example, sources | {brief description} |

### Reciprocal Link Maintenance

| Existing Concept | Link Type Added | New Concept |
|-----------------|----------------|-------------|
| {id} | related | {new_id} |

### Quality
- Orphan concepts: 0
- Cross-cluster connections: {count}
- Evidence tier distribution: {T1}% / {T2}% / {T3}%
- Counter-perspective ratio: {ratio}%

Step 10b: Update Registry

If NOT --dry-run and project registry exists:

1. Append project entry to registry
2. Update metadata.next_available_id
3. Update metadata.total_allocated

If registry does not exist, WARN and skip.

Step 10c: Expansion Retrospective

After reporting results and updating registry, prompt:

Expansion Retrospective (optional)

• Went well: What worked in this expansion cycle?
• Could improve: What would you change next time?
• Concepts: Any concepts that surprised you or seem off?

If user provides a response, include it in the session file under ## Retrospective. If user declines (empty response or "skip"), proceed without recording.

This step is non-blocking — the expansion is already complete and committed. The retrospective captures process learning for future cycles.

Constraints

• C1: NEVER fabricate sources — all sources MUST be real, verifiable references
• C2: NEVER create orphan concepts — all concepts MUST have >=1 related link to existing ontology
• C3: NEVER remove or modify existing concepts — expansion is additive only
• C4: NEVER skip web research unless --skip-web-search or --from-kb explicitly passed
• C5: NEVER write to vocabulary without running validation afterward
• C6: MUST log expansion to quality_history
• C7: MUST achieve 60%+ concepts at evidence Tier 1-2
• C8: MUST check for duplicates against existing prefLabels and altLabels
• C9: MUST present expansion areas to user before executing web research
• C10: MUST read current ontology state before proposing expansions
• C11: Enhancement is additive only — NEVER remove existing field content
• C12: Enhanced concepts MUST retain all existing field content
• C13: MUST verify prefLabels against PL-01 through PL-05 before writing; >10% failure triggers review
• C14: MUST update project registry after execution-mode expansion (if registry exists)
• C15: MUST verify 20-30% counter-perspective ratio before SKOS writing
• C16: MUST NOT contain hardcoded vocabulary, validation tool, or index paths — all from config block
• C17: MUST verify and maintain reciprocal SKOS links for all new concepts
• C18: MUST prompt for expansion provenance when --provenance is not provided; provenance is optional but prompted (not silently skipped)

Error Handling

Related Skills

• /aget-record-observation — Record raw observations before expansion
• /aget-record-lesson — Record expansion learnings
• /aget-study-topic — Research topic context before expansion
• /aget-analyze-ontology — Analyze ontology health (complementary)

Traceability

aget-expand-ontology v2.1.0 Category: Creation Merged: professional-core v1.5.0 + framework-AGET v1.0.0 + supervisor additions Spec: SKILL-019 (31 capabilities, 16 constraints) v2.1.0: +--provenance (EO-030, C18, #388) +Step 10c retro (EO-031, #561)