Context Store

Context Store Skill

What & When

What: Commands for uploading, downloading, querying and deleting documents with metadata and tags through the Context Store server. Also linking to one another via relations.

When to use:

• Store important documents for retrieval in future sessions
• Build a knowledge base of project documentation
• Share architecture, API specs, or design documents across different work sessions
• Query documents by tags to find relevant information

Key Benefits:

• Cross-session persistence
• Tag-based organization with AND logic
• Simple JSON output

Quick Reference

CRITICAL: Always use absolute paths - NEVER use cd:

doc-push - Upload Documents

uv run --script <skill-root>/commands/doc-push <file> [--tags TEXT] [--description TEXT]
# Example: uv run --script <skill-root>/commands/doc-push specs.md --tags "api,v2"

Use when: Store a document for future reference.

doc-query - Search by Tags/Name

uv run --script <skill-root>/commands/doc-query [--tags TEXT] [--name TEXT] [--include-relations]
# Example: uv run --script <skill-root>/commands/doc-query --tags "api,v2"
# With relations: uv run --script <skill-root>/commands/doc-query --tags "api" --include-relations

Use when: Find documents by tags (AND logic) or name patterns. Option: --include-relations adds relations to each document in response.

doc-search - Semantic Search

uv run --script <skill-root>/commands/doc-search "<query>" [--limit INT] [--include-relations]
# Example: uv run --script <skill-root>/commands/doc-search "how to configure authentication"
# With relations: uv run --script <skill-root>/commands/doc-search "auth" --include-relations

Use when: Find documents by meaning using natural language queries. Note: Requires semantic search enabled on server. Returns section offsets for partial reads. Option: --include-relations adds relations to each result.

doc-info - Get Document Metadata & Relations

uv run --script <skill-root>/commands/doc-info <document-id>
# Example: uv run --script <skill-root>/commands/doc-info doc_abc123

Use when: View metadata and relations for a specific document without downloading it. Output: Includes relations field with parent/child/related document links.

doc-read - Read Text Documents

uv run --script <skill-root>/commands/doc-read <document-id> [--offset INT] [--limit INT]
# Example: uv run --script <skill-root>/commands/doc-read doc_abc123
# Partial: uv run --script <skill-root>/commands/doc-read doc_abc123 --offset 2000 --limit 1000

Use when: Output text document content directly to stdout (text files only). Partial reads: Use --offset and --limit to retrieve specific sections (useful with semantic search results).

doc-pull - Download Documents

uv run --script <skill-root>/commands/doc-pull <document-id> [-o PATH]
# Example: uv run --script <skill-root>/commands/doc-pull doc_abc123 -o specs.md

Use when: Retrieve a document by its ID.

doc-delete - Remove Documents

uv run --script <skill-root>/commands/doc-delete <document-id>
# Example: uv run --script <skill-root>/commands/doc-delete doc_abc123

Use when: Permanently remove a document.

doc-link - Manage Document Relations

uv run --script <skill-root>/commands/doc-link --types                     # List relation types
uv run --script <skill-root>/commands/doc-link --create <from> <to> [opts] # Create relation
uv run --script <skill-root>/commands/doc-link --update <id> --note "..."  # Update note
uv run --script <skill-root>/commands/doc-link --remove <id>               # Remove relation

Use when: Link documents together in parent-child or peer relationships. Options for --create:

• --type - Required: parent-child (hierarchical) or related (peer)
• --from-to-note - Note on edge from source to target (source's note about target)
• --to-from-note - Note on edge from target to source (target's note about source)

Typical Workflows

Store and Retrieve

# Upload with tags
uv run --script <skill-root>/commands/doc-push specs.md --tags "api,v2"

# Find it later
uv run --script <skill-root>/commands/doc-query --tags "api,v2"

# Download it
uv run --script <skill-root>/commands/doc-pull doc_abc123

Build Knowledge Base

# Upload multiple documents with consistent tags
uv run --script <skill-root>/commands/doc-push architecture.md --tags "design,mvp"
uv run --script <skill-root>/commands/doc-push api-spec.md --tags "api,mvp"

# Query by project phase
uv run --script <skill-root>/commands/doc-query --tags "mvp"

Semantic Search + Partial Read

# Search by meaning
uv run --script <skill-root>/commands/doc-search "how to authenticate users"
# Returns: {"results": [{"document_id": "doc_abc", "sections": [{"offset": 2000, "limit": 1000}]}]}

# Read only the relevant section
uv run --script <skill-root>/commands/doc-read doc_abc --offset 2000 --limit 1000

Link Related Documents

# Get relation types
uv run --script <skill-root>/commands/doc-link --types

# Create a sequential relationship
uv run --script <skill-root>/commands/doc-link --create doc_design doc_implementation --type predecessor-successor --from-to-note "Design doc" --to-from-note "Implementation plan"

# Create hierarchical relationship (parent owns children)
uv run --script <skill-root>/commands/doc-link --create doc_architecture doc_api --type parent-child --from-to-note "API spec"

# Create peer relationship
uv run --script <skill-root>/commands/doc-link --create doc_api doc_models --type related --from-to-note "Data models"

# View document with its relations
uv run --script <skill-root>/commands/doc-info doc_architecture

Key Concepts

Tag AND Logic

IMPORTANT: Multiple tags means ALL must match:

• --tags "python,api" → Document must have BOTH tags
• --tags "v2,design,spec" → Document must have ALL THREE tags

Output Format

All commands output JSON. Save document IDs from upload for later retrieval/deletion.

Document Relations

Relation Types:

• parent-child: Hierarchical ownership. Deleting parent cascades to children.
• related: Peer relationship. No cascade delete.

Bidirectional: Relations are stored from both perspectives. Creating a parent-child link adds:

• parent relation on the parent document
• child relation on the child document

Notes for AI Assistants

1. Tag AND logic - Multiple tags = ALL must match
2. Save document IDs - From upload output for future operations
3. Check server running - Handle connection errors gracefully
4. Parse JSON output - All commands return JSON
5. Tags are lowercase - Use consistent tag naming (python not Python)
6. Relations in doc-info - Use doc-info to see document relations without a separate call
7. Relation IDs - Save relation IDs from create output for update/remove operations
8. Cascade delete - Deleting a parent document also deletes its children

Quick Decision Tree

Store document? → doc-push <file> --tags "tag1,tag2"

Find by tags? → doc-query --tags "tag1,tag2" (AND logic, add --include-relations for links)

Find by meaning? → doc-search "your question" (add --include-relations for links)

Check metadata? → doc-info <doc-id> (metadata + relations)

Read text file? → doc-read <doc-id> (text files to stdout)

Read section? → doc-read <doc-id> --offset 2000 --limit 1000 (partial read)

Download document? → doc-pull <doc-id> (ID from query)

Remove document? → doc-delete <doc-id> (permanent)

List all? → doc-query (no args)

Link documents? → doc-link --create <from> <to> --type parent-child

List relation types? → doc-link --types

Update link note? → doc-link --update <id> --note "text"

Remove link? → doc-link --remove <id> (removes both directions)

Additional Resources

• Detailed Command Reference: See references/COMMANDS.md
• Configuration Options: See references/CONFIGURATION.md
• Error Handling: See references/TROUBLESHOOTING.md