vocion-core

List every agent, skill, and object type defined in the active workspace directory. Returns slugs, names, and descriptions — not full prompts.

Return the full manifest (including prompt text) for one agent, skill, or object type. Use after workspace_list to pick the slug.

Write a skill SKILL.md (frontmatter + body) to workspace/<org>/skills/. Writes to disk + auto-applies to DB. Git is external — pass autoCommit=true to opt in. Returns files written, new context SHA, and the apply diff.

Write an agent manifest + system prompt. Writes to disk + auto-applies to DB. Git is external — pass autoCommit=true to opt in.

Write an object type manifest (schema, source relevance, classification prompt). Writes to disk + auto-applies to DB. Git is external — pass autoCommit=true to opt in.

Remove the files for an agent, skill, or object type from workspace/<org>/. Auto-commits + auto-applies (which will remove the row from the DB on the next apply).

Reconcile workspace/<org>/ to the database and record a workspace_version audit row. Use after editing files directly (outside MCP) or when auto-apply was disabled.

Dry-run apply — shows created/updated/unchanged counts without writing. Equivalent to `workspace_apply dryRun=true`.

Show the last N workspace_version rows: sha, summary, applied_by, applied_at. Useful for answering "when did this prompt last change?"

Return business_object rows for this org. Filter by type slug (e.g. discovery_call, deal).

Return a business_object row with its type and document links.

Return the business_object_type rows. Useful for discovering what type slugs are valid for objects_list.

Run a retrieval query against indexed documents (pgvector + Postgres FTS hybrid). Returns ranked documents with snippets, source type, dates, and links. Filter by sources (e.g. ["zoom","gmail"]) to narrow scope.

Live web search. Returns ranked titles, URLs, and snippets.

Fetch one URL and return its readable text/markdown.

Same-origin BFS crawl; returns pages (capped depth + count).

Generate an image from a prompt; returns the saved artifact URL.

Evaluate a math expression (builtin) or run code (sandbox provider).

Create a CSV, SVG chart, or doc and return its URL.

List authored mission templates (starting points) for this org.

Start an open-ended team mission from a brief — from a template (missionSlug) or ad-hoc (team). Plans the work, then runs to completion or the first approval gate.

List mission runs (optionally filtered by status).

Get one mission run with its plan (task graph), team, artifacts, and status.

Approve a paused mission run (awaiting_review) and continue execution.

Cancel a mission run.

Draft a reusable Workflow from a completed mission run (for review before activating).

Return the workspace lead plus every team of agents: lead agent, member roster, and exactly one resolved accountable human per team. Accountability provenance rides on `accountable.source` — "team" = set explicitly on the team, "workspace" = inherited from the workspace default. A null lead means "n…

Return every skill registered via the plugin system (separate from prompt-only skills in the DB).

Clear the registry and re-run plugin discovery. Use during local development after updating a plugin package. Hot-reload of edited local files is limited by Node ESM caching.

Return every workflow defined in workspace/<org>/workflows/ with its trigger and step count.

Return the complete manifest for a workflow including all step bodies.

Kick off a workflow manually. If the workflow hits an approve step, the run pauses and returns with status=paused.

Return recent workflow_run rows, newest first. Filter by workflow slug or status.

Full detail for one run including per-step results.

After a human approves the current approve step, continue execution from where it left off.

Permanently cancel a running or paused run. Cannot be un-cancelled.

Return the catalog of skills and playbooks for this org (slug, name, description, kind, origin, version, license).

Return a playbook's SKILL.md body and optionally one named sibling resource (REFERENCE.html, etc.). Resources are listed on the catalog row.