Initialize Cortex project. You MUST ask the human ALL 10 onboarding questions FIRST using cortex_ask_human. Do NOT call this tool until you have all answers. The 10 mandatory questions are: 1) Project name, 2) Goal, 3) Tech stack, 4) UI style, 5) Preferred libs, 6) Forbidden libs, 7) File structure, 8) Existing codebase, 9) Core features, 10) Complexity level.

Get compressed project state (under 200 tokens) — call before every action. Includes token budget warning if >50% used.

Get full dictionary entry for a file or feature

Get next pending todo - agent always calls this, never decides own next step

Log a completed step

Add a new feature

Update feature status

Mark a file as created/done in file tree. Saves checkpoint if file exists on disk.

Add a test linked to a feature

Pass or fail a test

Log a bug or blocker

Mark issue resolved with fix description

Write or update a dictionary entry

Save a code snippet to library

Save library research notes

Log an architectural decision

Run all 6 reminder checks — returns list of warnings/blocks based on real DB state

Compress current session to summary, save to DB and .cortex/snapshots/ on disk

Restore file to checkpoint

Force pause, queue a question for human, block until answered via POST /api/human-answer

Block dangerous actions until human confirms. Shows red modal in dashboard. Falls back to terminal if no dashboard.

Check DB integrity, missing tests, open issues, broken files. Auto-repairs common problems.

Switch active project (loads different .cortex DB). Blocked when process was started with --project (locked to single project).

List all available Cortex skills — bundled with the package (always available) plus any user-added skills. Returns name, source ("bundled" or "user"), and description for each.

Get the full content of a Cortex skill file. Call this at the start of any session to load the complete agent guide. Works with Claude Code, Gemini, Codex, Cursor, OpenCode, and any MCP client. Defaults to the main "cortex" skill (SKILL.md) if no name is given.

Search across dictionary, issues, snippets, research, and progress. Supports keyword (FTS5), semantic (vector), and hybrid (RRF fusion) modes. Returns ranked results with source table, id, and matched snippet.

Find conflicting data in the dictionary. If key provided, checks only that entry. If no key, scans recent entries.

Resolve a contradiction by choosing which version to keep.

Log token usage for the current action. Call after every tool call.

Get token budget status for the current session.

Set the role for an agent. Roles control which tools the agent can call.

Get the role and permissions for an agent.

List all connected agents and their roles.

Create a relationship between two entities (feature, file, test, issue, decision).

Start an autonomous pipeline from a goal. Breaks goal into tasks, executes them, auto-retries failures.

Get the status of a running or completed pipeline.

Pause the currently running pipeline.

Resume a paused pipeline.

Cancel a running or paused pipeline.

Preview what tasks a pipeline would create for a goal without starting it.

Auto-detect and fix broken files, failed tests, and other issues. Returns list of healed items.

View recent pipeline runs and their results.