hiivmind-corpus-graph

name	hiivmind-corpus-graph
description	View, validate, and edit a corpus concept graph (graph.yaml). Triggers: "graph", "show graph", "validate graph", "concept graph", "view concepts", "graph.yaml", "show relationships", "add concept", "add relationship".

Corpus Concept Graph

View, validate, and edit the concept graph (graph.yaml) for a corpus.

Precondition

A corpus must exist with a graph.yaml file. If no graph exists, inform the user: "This corpus does not have a concept graph yet. Run build with extraction enabled to generate one, or use add-concept to create one manually."

---

Phase 1: Locate Graph

Inputs: Corpus path (from user or auto-detected)

Procedure:

1. Locate the corpus using discovery patterns
2. Check for graph.yaml alongside index.md
3. Load and parse the graph

See: ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/discovery.md See: ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/graph.md

---

Phase 2: Execute Subcommand

Determine which subcommand to run based on user intent:

show

Display a human-readable summary of the concept graph.

Procedure:

1. List all concepts with: label, entry count, tags
2. List all relationships with: from → to (type), origin
3. Show meta: generated_at, total counts
4. Format as a readable table or structured list

Output example:

Concepts (8):
  family-activities    — 5 entries, tags: [family]
  recipes              — 3 entries
  work-projects        — 3 entries, tags: [project]
  ...

Relationships (5):
  family-activities → recipes (includes) [wikilink]
  work-projects → budget-review (see-also) [manual]
  ...

Generated: 2026-03-13T00:00:00Z | 24 entries | 8 concepts | 5 relationships

validate

Check graph.yaml for issues.

Procedure:

1. Load graph.yaml and index.md
2. Run all validation rules from ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/graph.md § Validation Rules
3. Report: errors (must fix), warnings (should review)
4. If clean: "Graph validates OK — N concepts, M relationships, no issues."

Headless mode (validate --headless, or headless: true input): run the same validation rules non-interactively and write {corpus_path}/graph-validate-result.yaml (headless contract — see ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/headless-contract.md):

contract_version: 1
kind: graph-validate
corpus: {name}
run_at: {ISO 8601}
concepts: {int}
relationships: {int}
issues:
  - severity: error | warning
    rule: {rule id from patterns/graph.md § Validation Rules}
    detail: "{human-readable description with the offending id}"
valid: {bool}                    # true when zero error-severity issues
errors: []                       # runtime failures, not validation findings

Exit semantics for pipelines: emit the file always; valid: false when any error-severity issue exists. Validate with uv run ${CLAUDE_PLUGIN_ROOT}/lib/corpus/scripts/validate_result.py graph-validate-result.yaml --kind graph-validate. Ensure graph-validate-result.yaml is gitignored in the corpus (append if missing). Note: interactive rule 2 loads index.md; headless mode loads index.yaml when present (v2), since entry IDs are the join key being checked.

add-concept

Add a new concept to graph.yaml with semi-automated entry suggestion.

Invocation: graph add-concept "Concept Label" or interactive (no args → prompt for label)

Procedure:

1. Slugify label to concept ID per ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/graph.md § Concept ID Conventions:

   slugify() {
     echo "$1" | tr '[:upper:]' '[:lower:]' | sed 's/[^a-z0-9]/-/g; s/--*/-/g; s/^-//; s/-$//'
   }
   

2. Check for duplicates — if concept ID already exists in graph.yaml, error: "Concept '{id}' already exists. Use a different name or edit the existing concept."
3. Scan index for matching entries:
   • Extract 2-4 search terms from the concept label
   • v2 (index.yaml): yq pre-filter on tags, keywords, summary fields. If yq unavailable, read index.yaml directly and use LLM judgment.

     yq '.entries[] | select(
       (.tags[] | test("term1|term2"; "i")) or
       (.keywords[] | test("term1|term2"; "i")) or
       (.summary | test("term1.*term2|term2.*term1"; "i"))
     ) | {id, title, summary, tags}' index.yaml
     

   • v1 (index.md): grep for search terms + LLM semantic judgment
   • Present top 10-15 candidates ranked by relevance
4. User selects entries — present numbered list, user confirms/deselects
5. Suggest tags — based on selected entries' metadata (if available from index.yaml)
6. Suggest description — one-line summary based on selected entries' summaries
7. Write concept to graph.yaml (v2 schema — no entry lists):

   {concept-id}:
     label: "{User-Provided Label}"
     description: "{description}"
     tags: ["{tag1}", "{tag2}"]
   

8. Update index.yaml entries with concept reference: For each selected entry, update its concepts array in index.yaml:
   • Read index.yaml
   • Find entry by ID

Empty graph bootstrap: If graph.yaml doesn't exist, create a new one:

schema_version: 2

concepts:
  {concept-id}:
    label: "{label}"
    description: "{description}"
    tags: []

relationships: []

meta:
  generated_at: "{ISO-8601}"
  concept_count: 1
  relationship_count: 0

graph.yaml v1 compatibility: If existing graph.yaml has schema_version: 1 (concepts with entries[] lists), migrate on edit: move entry lists to index.yaml concepts[], remove entries and entry_count from graph.yaml, set schema_version: 2.

Output: "Added concept '{label}' with {n} entries to graph.yaml and index.yaml"

add-relationship

Add a typed relationship between two existing concepts. Supports explicit (args) and interactive (no args) modes.

Explicit invocation: graph add-relationship "{from-concept}" "{to-concept}" {type}

Procedure (explicit):

1. Validate concept IDs — both from and to must exist in graph.yaml. If not: "Concept '{id}' not found. Available concepts: {list}"
2. Validate relationship type — must be one of the controlled vocabulary: includes, depends-on, see-also, extends, contrast-with (see ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/graph.md § Relationship Type Vocabulary)
3. Check for duplicates — same from/to/type triple already exists: "Relationship '{from}' → '{to}' ({type}) already exists."
4. Optionally ask for evidence — an entry path that demonstrates the link (can be null)
5. Write relationship to graph.yaml:

   - from: "{from-concept}"
     to: "{to-concept}"
     type: "{type}"
     origin: "manual"
     evidence: null
   

6. Update meta — increment relationship_count, set generated_at

Interactive invocation (no args):

1. Load graph.yaml, list all concepts
2. Analyze concept pairs for relationship candidates:
   • Tag overlap: Concepts sharing 2+ tags → strong candidate
   • Entry cross-references: If entries in concept A link to entries in concept B
   • Label/description keyword similarity: Fuzzy match on descriptions
   • Embedding similarity (if index-embeddings.lance/ exists and fastembed available): For each concept, query index-embeddings.lance/ with the concept's label+description. Concepts with entries that are semantically similar (score > 0.7) but not yet linked are relationship candidates.
3. Present candidates:

   These concept pairs look related — confirm or skip each:
   
   1. query-optimization → lazy-frames
      Reason: shared tags [performance, optimization]
      Suggested type: depends-on
      [Confirm / Change type / Skip]
   
   2. data-types → serialization
      Reason: entries cross-reference each other
      Suggested type: see-also
      [Confirm / Change type / Skip]
   

4. For each confirmed pair, write relationship with origin: "manual"
5. Summary: "Added {n} relationships to graph.yaml"

Note: All relationships added by this subcommand use origin: "manual", which means refresh will preserve them (never overwritten by auto-extraction).

---

Error Handling

Error	Message	Recovery
No corpus found	"Could not locate a corpus. Specify the path or run from a corpus directory."	Ask user for path
No graph.yaml	"This corpus has no concept graph. Generate one with build (extraction enabled)."	Suggest build
Invalid YAML	"graph.yaml has syntax errors: {details}"	Show error location
Schema violation	"graph.yaml schema issue: {details}"	Show specific field

---

Pattern Documentation

• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/graph.md — Graph schema, validation rules, concept ID conventions
• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/discovery.md — Corpus location discovery
• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/config-parsing.md — Config.yaml parsing
• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/embeddings.md — Embedding-based similarity for relationship candidates

Related Skills

• Migrate v1→v2 (headless): ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-migrate/SKILL.md
• Headless status (pipelines): ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-status-headless/SKILL.md
• Headless rebuild (pipelines): ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-build-headless/SKILL.md
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-build/SKILL.md — Generates graph.yaml during build
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-navigate/SKILL.md — Uses graph.yaml for enriched retrieval
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-bridge/SKILL.md — Cross-corpus concept bridges and aliases
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-enhance/SKILL.md — Can add entries to concepts
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-refresh/SKILL.md — Updates auto-extracted relationships
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-add-source/SKILL.md — Configures extraction
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-init/SKILL.md — Initializes corpus
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-discover/SKILL.md — Finds installed corpora
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-register/SKILL.md — Registers corpus to project
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-status/SKILL.md — Reports corpus health