Skip to content

PROMPT.md

Latest commit

 

History

History
165 lines (131 loc) · 8.28 KB

PROMPT.md

File metadata and controls

165 lines (131 loc) · 8.28 KB

Graph Memory — AI Assistant Prompt

Copy the section below into your CLAUDE.md, .cursorrules, or equivalent AI assistant configuration file.

Graph Memory

You have access to Graph Memory — an MCP server that maintains a semantic graph of this project. Use it as your primary source of context before reading files directly.

When to use

  • Before diving into code — search first, read files second. Graph Memory already indexed the codebase and docs.
  • To find relevant code — search_code finds symbols by meaning, not just name. Use it instead of grepping.
  • To find documentation — search finds relevant doc sections even if the wording differs from your query.
  • To understand a symbol — get_symbol returns full source, explain_symbol returns docs + examples, cross_references bridges code ↔ docs.
  • To track decisions — create notes in the knowledge graph so context persists across conversations.
  • To manage work — create and move tasks through the kanban workflow.
  • To store recipes — save reusable procedures as skills so they can be recalled later.

Docs tools

Search and browse indexed markdown documentation.

Tool Purpose
search Hybrid search (BM25 + vector) over doc sections with BFS expansion
search_topic_files File-level semantic search over docs (by file path + title)
list_topics List all indexed markdown files with title and chunk count
get_toc Table of contents for a specific doc file
get_node Full content of a specific doc section by ID

Code block tools

Search code examples extracted from documentation.

Tool Purpose
find_examples Find code blocks in docs containing a specific symbol
search_snippets Semantic search over code blocks extracted from docs
list_snippets List code blocks with filters (file, language, content)
explain_symbol Find code example + surrounding text explanation for a symbol

Cross-graph tools

Tool Purpose
cross_references Full picture for a symbol: code definition + doc examples + explanations

Code tools

Search and browse indexed TypeScript/JavaScript source code.

Tool Purpose
search_code Hybrid search (BM25 + vector) over code symbols with BFS expansion
search_files File-level semantic search over source files (by path)
list_files List all indexed source files with symbol counts
get_file_symbols List all symbols in a source file (sorted by line)
get_symbol Full source body of a specific symbol by ID

File index tools

Browse all project files (not just docs/code pattern-matched).

Tool Purpose
search_all_files Semantic search over all files by path
list_all_files List all files/dirs with filters (directory, extension, language)
get_file_info Full metadata for a file or directory

Knowledge tools

Persistent knowledge graph for facts, decisions, and notes. Notes are mirrored to .notes/{id}/note.md.

Tool Purpose
create_note Create a note with title, content, and tags
update_note Update note title, content, or tags
delete_note Delete a note and all its relations
get_note Fetch a note by ID
list_notes List notes with optional filter and tag
search_notes Hybrid search over notes with BFS expansion
create_relation Create relation between notes, or to doc/code/file/task nodes
delete_relation Delete a relation (note-to-note or cross-graph)
list_relations List all relations for a note (includes cross-graph links)
find_linked_notes Reverse lookup: find notes linked to a doc/code/file/task node
add_note_attachment Attach a file to a note
remove_note_attachment Remove an attachment from a note

Example:

create_note({ title: "Why we chose PostgreSQL", content: "...", tags: ["architecture", "database"] })
create_relation({ fromId: "why-we-chose-postgresql", toId: "src/db/connection.ts", targetGraph: "code", kind: "documents" })

Task tools

Kanban task management with workflow: backlog → todo → in_progress → review → done | cancelled. Tasks are mirrored to .tasks/{id}/task.md.

Tool Purpose
create_task Create a task with title, description, priority, status, tags, dueDate, estimate
update_task Update any task fields (partial update)
delete_task Delete a task and all its relations
get_task Fetch task with subtasks, blockedBy, blocks, and related
list_tasks List tasks with filters (status, priority, tag, text)
search_tasks Hybrid search over tasks with BFS expansion
move_task Change task status (auto-manages completedAt)
link_task Create task↔task relation (subtask_of, blocks, related_to)
create_task_link Link task to a doc/code/file/knowledge node
delete_task_link Remove a cross-graph link from a task
find_linked_tasks Reverse lookup: find tasks linked to a target node
add_task_attachment Attach a file to a task
remove_task_attachment Remove an attachment from a task

Example:

create_task({ title: "Fix auth redirect loop", description: "...", priority: "high", status: "todo" })
move_task({ taskId: "fix-auth-redirect-loop", status: "in_progress" })
link_task({ fromId: "fix-auth-redirect-loop", toId: "write-auth-tests", kind: "blocks" })
create_task_link({ taskId: "fix-auth-redirect-loop", targetId: "src/auth.ts::login", targetGraph: "code", kind: "fixes" })

Skill tools

Reusable recipes, procedures, and troubleshooting guides. Skills are mirrored to .skills/{id}/skill.md.

Tool Purpose
create_skill Create a skill with steps, triggers, tags, and metadata
update_skill Update any skill fields (partial update)
delete_skill Delete a skill and all its relations
get_skill Fetch skill with dependsOn/dependedBy/related/variants
list_skills List skills with filters (source, tag, text)
search_skills Hybrid search over skills with BFS expansion
recall_skills Recall relevant skills for a task context (higher recall)
bump_skill_usage Increment usage counter + set lastUsedAt
link_skill Create skill↔skill relation (depends_on, related_to, variant_of)
create_skill_link Link skill to a doc/code/file/knowledge/task node
delete_skill_link Remove a cross-graph link from a skill
find_linked_skills Reverse lookup: find skills linked to a target node
add_skill_attachment Attach a file to a skill
remove_skill_attachment Remove an attachment from a skill

Example:

create_skill({ title: "Add REST endpoint", description: "...", steps: ["1. Create route file", "2. Add Zod schema", "3. Register in router"], triggers: ["new endpoint", "new API route"], tags: ["api"] })
recall_skills({ query: "add a new API endpoint" })
bump_skill_usage({ skillId: "add-rest-endpoint" })
link_skill({ fromId: "add-rest-endpoint", toId: "debug-authentication-issues", kind: "related_to" })
create_skill_link({ skillId: "add-rest-endpoint", targetId: "src/api/rest/index.ts", targetGraph: "code", kind: "references" })

Best practices

  1. Search before reading files — search_code and search are faster and more targeted than reading files manually.
  2. Create notes for decisions — when you make an architectural choice or discover something non-obvious, save it.
  3. Link everything — link notes to code, tasks to files, skills to docs. Cross-graph links make the knowledge base navigable.
  4. Use recall_skills before complex tasks — there might already be a saved recipe.
  5. Use move_task instead of update_task for status changes — it manages completedAt automatically.
  6. Use cross_references to get the full picture — it bridges code definitions with documentation examples.
  7. Bump skill usage after applying a skill — it helps surface frequently used recipes.
  8. Use find_linked_tasks when working on a file — see what tasks are related before making changes.
  9. Attach files to notes/tasks/skills for screenshots, logs, diagrams, or any supporting artifacts.