Slash commands and skills

Cortex Hub ships three Claude Code commands. /cs and /ce are written into each project by the installer; /install is a global skill. The cortex-bar mod adds two more, listed in Mods.

Command	Where it lives	What it does
/cs	.claude/commands/cs.md	Opens a session and recalls context
/ce	.claude/commands/ce.md	Runs the gates, stores what was learned, closes the session
/install	~/.claude/skills/install/SKILL.md	Runs the installer in the current project
/cortex-bar, /cortex-calls	cortex-bar mod	Status bar and call pane; see Mods

/cs: start a session

1. cortex_session_start(repo, mode, agentId, ide, branch, clientSessionId). Keeps session_id, the project id and project.orgId.
2. cortex_knowledge_search and cortex_memory_search, in parallel.
3. cortex_changes: unseen commits and files from other agents.
4. cortex_task_pickup: Conductor tasks for this agent, if any.
5. git status; on a dirty tree, cortex_detect_changes with git diff HEAD.
6. A situational summary.
7. The workflow for the rest of the session:
   • score non-trivial plans with cortex_plan_quality (8.0 to pass, up to three tries);
   • find code with the lookup table, search once and read all ten hits;
   • search knowledge and memory on an error before fixing it;
   • cortex_detect_changes before a commit, cortex_code_reindex after a push;
   • for a Conductor task: accept, set in_progress, then complete it.

In Claude Code the hooks block edits until steps 1 and 2 have run.

INFO

Step 1 of /cs warns and runs git pull when the start response reports recentChanges. The current cortex_session_start does not return that field, so this branch does not fire; cortex_changes in step 3 covers the same ground.

/ce: end a session

1. cortex_detect_changes on the diff; warns when risk is HIGH.
2. pnpm build && pnpm typecheck && pnpm lint.
3. cortex_quality_report(gate_name: "Session Quality", ...).
4. Completes any Conductor task still in progress.
5. cortex_knowledge_store, when there is something the team should know.
6. cortex_memory_store with what was done, decisions and next steps.
7. cortex_session_end(sessionId, summary). The hub saves the summary as memory.
8. A final report.

/install

/install                       # set up this project
/install --check               # report what is installed
/install --force               # regenerate everything
/install --tools claude,gemini # only these IDEs

The skill downloads scripts/install.sh through the GitHub API, falling back to raw.githubusercontent.com, and runs it. It has disable-model-invocation: true, so Claude runs it only when you type it. Its source is templates/skills/install/SKILL.md.

In other IDEs

Cursor, Windsurf, VS Code Copilot and Codex have no slash commands from Cortex. Their rules files tell the agent to run the same steps through the MCP tools. See Multi-IDE.