STEP 1 of token-efficient exploration. Semantic search in code (OpenAI embeddings). Returns a lightweight index: file paths, line numbers, short previews — NOT full source. Then: call get_file_outline for file structure, get_file_folded for signatures+JSDoc, and get_symbol ONLY for the specific symbol you need. Never Read whole files after this — you have better tools.

Semantic search in documentation and markdown files. Finds relevant documentation sections by natural language query. Returns file paths, section headings, and full content.

Force incremental reindex of the project codebase. Only re-embeds files that have changed since last index. Use this to initialize the index for the first time. After first indexation, auto-index is enabled for subsequent startups via .stellarisrc. Use force=true after switching embedding providers (deletes the old index automatically).

Reindex a single file after it has been modified or created. Much faster than a full reindex — use this in hooks after Write/Edit tool calls to keep the index up-to-date in real time. Requires OPENAI_API_KEY.

Get the project file tree structure. Returns all indexed files organized by directory, with stats on languages and file counts. No API call needed — instant response.

STEP 2 of token-efficient exploration. Lists top-level symbols (functions/classes/types/hooks) with line ranges + imports/exports. ~200 tokens. After this, call get_file_folded for signatures+JSDoc, or get_symbol for full source of ONE symbol. NEVER Read the whole file — you have better tools. Uses AST, no API.

STEP 3 of token-efficient exploration. Returns all symbols with signatures + JSDoc but NO function bodies — folded view under a token budget. Ideal when you need to understand a file structurally without reading every implementation. Falls back to truncated:true if budget exceeded. AST-based, no API.

STEP 4 of token-efficient exploration — the only step that returns full source. Fetches ONE symbol (function/class/type) with file context (imports, siblings, warnings). Use ONLY after search_code/get_file_outline/get_file_folded identified the symbol. No API.

Get the files that a given file imports (its dependencies). Shows the dependency chain from this file outward. Requires a prior reindex to build the dependency graph. No API call needed.

Get the files that import a given file (its dependents / reverse dependencies). Shows what other code relies on this file. Requires a prior reindex to build the dependency graph. No API call needed.

Analyze the blast radius of changes to a file: find all files that would be directly or transitively affected. Uses BFS on the dependency graph. Returns severity assessment, impacted files by depth, and edges. Requires a prior reindex. No API call needed.

Get Claude Code token usage statistics and estimated API costs. Shows consumption by model, project, or day for a given period. No API key required.

Launch a local web dashboard showing Claude Code token usage with interactive charts. Opens in VS Code Simple Browser. No API key required.

List Claude Code sessions that hit health thresholds: SES001 cost ≥$25, SES002 ≥200 turns, SES003 ≥5M tokens, SES004 idle 7+ days. No API key required.

Introspect a database and save a local schema snapshot to .vectors/db-schema.json. Connects to the DB via a connection string (PostgreSQL supported), or falls back to parsing local schema files (prisma/schema.prisma, database.types.ts). Run this once before using db_schema or db_search.

Read the local database schema snapshot. Returns tables, columns, types, primary keys, foreign keys, indexes, enums, and RLS policies. No DB connection needed — reads from .vectors/db-schema.json. Run db_snapshot first to create or refresh the snapshot.

Search the database schema by natural language query. Finds tables and columns matching a concept (e.g. "user permissions", "image generation prompts", "article cover"). No DB connection needed — searches the local snapshot.

Launch a 3D interactive visualization of the project dependency graph. Shows files as colored nodes (by language) connected by import edges. Supports filtering by file type, focusing on a specific file neighborhood, and clicking nodes to inspect file details and open them in VS Code. Requires a prior reindex to build the dependency graph. Opens in VS Code Simple Browser.

Detect circular dependencies in the project using Tarjan's SCC algorithm. Returns groups of files that form dependency cycles. Use this before refactoring to identify problematic coupling. Requires a prior reindex. No API call needed.

Find files that are never imported by any other file (dead code candidates). Excludes known entry points (index, main, config, test files). Use this to identify unused code before cleanup. Requires a prior reindex. No API call needed.

Returns files in dependency order (dependencies before dependents). Use this to determine the safe order to modify files during a refactor — process files in this order to avoid breaking intermediate builds. Requires a prior reindex. No API call needed.

Simulate moving a file from one path to another. Returns which files need import updates and what the new import strings should be. Use this before any file rename or refactor to get a complete migration plan. Requires a prior reindex. No API call needed.

Returns the most highly coupled files (highest combined in-degree + out-degree). High coupling signals refactoring candidates. Files with many consumers (high in-degree) are risky to change; files with many imports (high out-degree) have broad dependencies. Requires a prior reindex. No API call needed.

Aggregated project health check. Runs cycle detection (Tarjan), dead code analysis, coupling hotspots, graph complexity stats, max import depth, and index freshness — returns a global score A–F. Use as a quick diagnostic before any major refactoring. No API call needed.

Condensed project briefing designed for Claude Code SessionStart hook. Returns: graph health summary, cycles, and recent git activity with blast-radius ranking — all under ~800 tokens. Degrades gracefully if no git or no graph. Pairs with nova-mind-cloud searchMemory for complete context priming.

Heuristic detector for "was this session technically significant?" Returns significant:true + signals if git diff exceeds thresholds (default: 100 lines or 5 files) or graph has cycles. Intended for Stop/SessionEnd hooks to prompt memorization via nova-mind-cloud storeMemory. No API.

Show where Claude Code tokens go: task category breakdown (coding, debugging, feature, refactoring, testing, exploration, planning, delegation, git, build_deploy, conversation, brainstorming), MCP server call counts, and core tool usage. Inspired by Codeburn. No API key required.

Export a static architecture diagram from the dependency graph. Groups files by semantic layer (Tools, Storage, Graph, Indexer, Analytics, API, Frontend, Backend, Security, Config) based on directory-name heuristics. Three output formats: mermaid (copy-paste in README, renders on GitHub/GitLab), svg (standalone dark-theme file), html (self-contained page with legend + download button). Requires a prior reindex. No API call needed.

Returns architecture boundary violations detected at index time. Rules are loaded from stellaris.boundaries.json at project root (format: { "deny": [{ "from": "src/ui/**", "to": "src/db/**", "reason": "..." }] }). Patterns are glob-style. Use this to enforce layering rules without runtime overhead. Requires a prior reindex.

Find documentation/markdown files that reference a code symbol or file (via backtick-quoted identifiers). Use this to discover where a symbol is documented, or to check if a file you are about to delete is mentioned in the docs. Requires a prior reindex.