hiivmind-corpus-bridge

Cross-Corpus Concept Bridge

Create and manage cross-corpus concept bridges stored in .hiivmind/corpus/registry-graph.yaml. Bridges link concepts that exist in different corpora, and aliases provide query-routing hints so searches can span multiple corpora.

Precondition

• .hiivmind/corpus/registry.yaml must exist with 2+ registered corpora
• At least 2 registered corpora must have graph.yaml files

If prerequisites not met:

• No registry: "No corpus registry found at .hiivmind/corpus/registry.yaml. Register corpora first: /hiivmind-corpus register"
• Fewer than 2 corpora: "Bridge requires at least 2 registered corpora. Found: {n}"
• No corpora with graph.yaml: "No registered corpora have concept graphs. Build graphs first with /hiivmind-corpus graph add-concept or run build with extraction."

---

Phase 1: Load Registry + Fetch Graphs

Inputs: Project working directory

Procedure:

1. Read .hiivmind/corpus/registry.yaml
2. For each registered corpus, fetch its graph.yaml:
   • Resolve source details from registry entry: source.repo (split on / for owner/repo) and source.ref
   • GitHub sources: gh api repos/{owner}/{repo}/contents/graph.yaml?ref={ref} --jq '.content' | base64 -d
   • Local sources: Read: {path}/graph.yaml
   • Embedded (self) sources: Read: .hiivmind/corpus/graph.yaml
   • Skip corpora without graph.yaml (report which were skipped)
3. Build in-memory map: corpus_name → { concepts, relationships }
4. Report: "Loaded graphs from {n} corpora: {names}. Skipped {m}: {names} (no graph.yaml)"

Pattern references:

• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/registry-loading.md
• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/registry-graph.md

---

Phase 2: Execute Subcommand

Determine which subcommand to run based on user intent:

bridge (default) — Interactive Creation

Detect bridge candidates and present for user confirmation.

Procedure:

1. Detect candidates using algorithm from ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/registry-graph.md § Bridge Candidate Detection: a. Label similarity — concepts with similar labels across corpora (shared words after removing stop words like "the", "and", "of") b. Tag overlap — concepts sharing tags across corpora c. Keyword overlap — corpus-level keywords that appear in another corpus's concept labels d. Embedding similarity (if corpora have index-embeddings.lance/ and fastembed available):
   • For each concept in corpus A's graph.yaml:
     • Search corpus B's index-embeddings.lance/: python3 search.py {B}/index-embeddings.lance/ "{concept.label} {concept.description}" --top-k 5 --select "concepts" --json
     • Read concepts from each result entry (returned by --select)
     • If entry belongs to a concept in corpus B → bridge candidate pair
   • Merge with candidates from label/tag/keyword matching (deduplicate)
2. Present candidates grouped by corpus pair:

   Between Polars and Ibis (3 candidates):
   
   1. polars:lazy-evaluation ↔ ibis:deferred-execution
      Match: embedding similarity 0.84, tag overlap [performance, lazy]
      Suggested type: see-also
      [Confirm / Change type / Skip]
   
   2. polars:data-types ↔ ibis:type-system
      Match: tag overlap [types, schema]
      Suggested type: see-also
      [Confirm / Change type / Skip]
   

3. Collect notes — for each confirmed bridge, optionally ask for a note
4. Suggest aliases — analyze confirmed bridge pairs for natural search terms:

   Suggested aliases:
   1. "lazy evaluation" → polars:lazy-evaluation, ibis:deferred-execution
   2. "type system" → polars:data-types, ibis:type-system
   [Confirm each / Skip all]
   

5. Write registry-graph.yaml (create or update):

   schema_version: 1
   
   bridges:
     - concept_a: "polars:lazy-evaluation"
       concept_b: "ibis:deferred-execution"
       type: "see-also"
       note: "Both implement deferred query execution for performance"
   
   aliases:
     "lazy evaluation":
       - corpus: "polars"
         concept: "lazy-evaluation"
       - corpus: "ibis"
         concept: "deferred-execution"
   
   meta:
     updated_at: "{ISO-8601}"
     bridge_count: 1
     corpora_linked: ["polars", "ibis"]
   

6. Confirm: "Created {n} bridges and {m} aliases in registry-graph.yaml"

Note: Cross-corpus semantic search uses each corpus's index-embeddings.lance/ directly. No separate registry-embeddings.lance/ is needed. Bridge candidate detection (step 1d) queries per-corpus embeddings and reads the concepts column to identify concept relationships.

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

show

Display existing bridges and aliases in human-readable format.

Procedure:

1. Read .hiivmind/corpus/registry-graph.yaml
2. If not found: "No cross-corpus bridges configured. Run bridge to create some."
3. Display:

   Cross-Corpus Bridges ({n}):
     polars:lazy-evaluation ↔ ibis:deferred-execution (see-also)
       "Both implement deferred query execution"
     polars:data-types ↔ ibis:type-system (see-also)
   
   Aliases ({m}):
     "lazy evaluation" → polars:lazy-evaluation, ibis:deferred-execution
     "type system" → polars:data-types, ibis:type-system
   
   Linked corpora: polars, ibis
   Last updated: {date}
   

validate

Check registry-graph.yaml for issues.

Procedure:

1. Read .hiivmind/corpus/registry-graph.yaml
2. Run validation rules from ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/registry-graph.md § Validation Rules: a. Schema compliance — all required fields present, correct types b. Corpus resolution — each corpus_name resolves to an installed corpus via registry.yaml c. Concept resolution — each concept-id exists in the referenced corpus's graph.yaml d. Dangling references — flag bridges where one side's corpus is not installed (warning, not error) e. Relationship type vocabulary — only controlled types used f. Alias targets — each alias target resolves to valid corpus + concept
3. Report errors (must fix) and warnings (should review)
4. If clean: "Registry graph validates OK — {n} bridges, {m} aliases, no issues."

add-alias

Manually add a query-routing alias.

Invocation: bridge add-alias "search term" corpus1:concept1 corpus2:concept2

Procedure:

1. Validate each corpus:concept reference resolves (corpus in registry, concept in its graph.yaml)
2. Check if alias key already exists — if so, offer to update: "Alias 'search term' already exists. Update it? [Yes / No]"
3. Write alias to registry-graph.yaml:

   aliases:
     "search term":
       - corpus: "corpus1"
         concept: "concept1"
       - corpus: "corpus2"
         concept: "concept2"
   

4. Update meta.updated_at
5. Confirm: "Added alias 'search term' → corpus1:concept1, corpus2:concept2"

---

Error Handling

Error	Message	Recovery
No registry	"No corpus registry found at .hiivmind/corpus/registry.yaml"	Suggest /hiivmind-corpus register
Fewer than 2 corpora	"Bridge requires at least 2 registered corpora. Found: {n}"	Suggest registering more
No corpora with graph.yaml	"No registered corpora have concept graphs. Build graphs first."	Suggest /hiivmind-corpus graph add-concept or build with extraction
Invalid registry-graph.yaml	"registry-graph.yaml has syntax errors: {details}"	Show error location
Network error fetching graph	"Could not fetch graph.yaml for corpus '{id}': {error}"	Skip that corpus, continue with others
Dangling concept reference	"Concept '{id}' not found in corpus '{name}' graph.yaml"	Warning during validate, not blocking

---

Pattern Documentation

• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/registry-graph.md — Registry graph schema, validation rules, bridge detection
• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/graph.md — Per-corpus graph schema (concepts, relationships)
• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/registry-loading.md — Registry loading patterns
• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/corpus-routing.md — Query routing (aliases enhance this)
• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/discovery.md — Corpus location discovery
• ${CLAUDE_PLUGIN_ROOT}/lib/corpus/patterns/embeddings.md — Embedding generation for cross-corpus concepts

Related Skills

• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-graph/SKILL.md — Per-corpus concept graph (view, validate, edit)
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-navigate/SKILL.md — Uses bridges for Tier 4 cross-corpus retrieval
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-register/SKILL.md — Registers corpora to project registry
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-discover/SKILL.md — Finds installed corpora
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-status/SKILL.md — Reports corpus health
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-build/SKILL.md — Generates graph.yaml during build
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-refresh/SKILL.md — Updates auto-extracted relationships
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-enhance/SKILL.md — Deepens coverage on topics
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-init/SKILL.md — Initializes corpus
• ${CLAUDE_PLUGIN_ROOT}/skills/hiivmind-corpus-add-source/SKILL.md — Adds documentation sources