Skip to main content
These are Verify beta features, available to the closed beta cohort only. They may change or move out of the core MCP without notice.
Verify skills are slash commands you invoke in your agent conversation to load a focused set of instructions, tool catalogs, or multi-step workflows into the agent’s context. Each skill is scoped to a specific part of the Verify tooling — live browser, proof documents, Sightmap authoring, tunnels, or comments — so you load only what you need for the task at hand.

​
Workflow Skills vs. Atomic Skills

Verify skills come in two flavors: Workflow skills orchestrate a complete, multi-step process. They instruct the agent to follow a defined sequence of tool calls — capturing state before changes, making changes, capturing state after, and finalizing records. Use them when you want the agent to run an entire procedure end-to-end. Atomic skills are reference catalogs and usage guides for a specific group of tools. They teach the agent everything it needs to know about a tool group (e.g. all live-* tools, or all doc-* tools) so it can use them correctly on demand. Use them when you need the agent to have detailed tool knowledge available during an open-ended task.

​
Skills Reference

​
/subtext:proof

Type: Workflow skill The inner loop for UI change verification. Invoke this skill when you want the agent to document a change with full before-and-after evidence. The skill instructs the agent to:
  1. Open a Proof Doc with doc-create
  2. Navigate to the relevant view in the live browser
  3. Capture a BEFORE screenshot and attach it with doc-attach
  4. Make the requested UI changes
  5. Capture an AFTER screenshot and attach it
  6. Finalize the record with doc-close
The resulting Proof Doc contains a permanent, evidence-backed narrative of exactly what changed and what it looks like before and after. Share the closed doc URL in your PR or ticket.
Use /subtext:proof as the outer frame for any task that involves modifying UI — it ensures the agent creates an evidence record automatically, without you having to remember to ask for screenshots.

​
/subtext:live

Type: Atomic skill Loads the complete catalog of hosted browser tools (live-*) and their usage patterns into the agent’s context. After invoking this skill, the agent knows:
  • How to open and manage connections and views (live-connect, live-view-*)
  • How to snapshot, screenshot, and inspect pages (live-view-snapshot, live-view-screenshot, live-view-inspect)
  • How to interact with elements using UIDs (live-act-*)
  • How to inspect network traffic and console output (live-net-*, live-log-*)
  • How to evaluate scripts and emulate devices (live-eval-script, live-emulate)
  • How to handle signals and tunnel URLs (live-signal, live-tunnel)
See the Live Tools reference for the full tool listing.

​
/subtext:sightmap

Type: Atomic skill Loads the Sightmap schema and authoring guidance into the agent’s context. Use this skill when you want the agent to create or maintain .sightmap/ YAML files that enrich browser snapshots with semantic component names, source paths, and memory notes. After invoking this skill, the agent understands:
  • The components, views, requests, and memory sections of the Sightmap schema
  • How CSS selectors map to component names and source file paths
  • How memory entries appear as [Guide] annotations at the top of every snapshot
  • How sightmap files are uploaded via the sightmap_upload_url returned by live-connect

​
/subtext:tunnel

Type: Atomic skill Loads tunnel setup instructions so the hosted browser can reach a localhost dev server. Use this skill at the start of any task where you need the cloud browser to connect to your local environment. After invoking this skill, the agent knows how to:
  • Use live-tunnel to allocate a connection and retrieve the tunnel relay URL
  • Configure the subtext-tunnel local MCP server (@subtextdev/subtext-cli) so the relay bridges to your dev server
  • Use tunnel-connect, tunnel-disconnect, and tunnel-status to manage the tunnel lifecycle
The subtext-tunnel MCP server is a separate local server — it is not part of the standard session-review setup. Add it to your MCP config before invoking this skill:

​
/subtext:docs

Type: Atomic skill Loads the Proof Doc tool catalog (doc-*) and seed templates for common work record types. Use this skill when you want the agent to create, update, and close Proof Docs without running the full /subtext:proof workflow — for example, when you need finer-grained control over what gets attached and when. After invoking this skill, the agent knows:
  • The full doc-* tool set (doc-create, doc-update, doc-attach, doc-close, doc-read, doc-diff, doc-list)
  • Seed markdown templates for common document types (bug fix, verification run, audit trail)
  • Best practices for naming attachment sections and structuring evidence
See the Proof Docs reference for the full tool listing and lifecycle details.

​
/subtext:comments

Type: Atomic skill Loads the comment tool catalog (comment-*) for agent-user collaboration during live sessions. Use this skill when you want the agent to read, post, or respond to comments as part of an interactive task — letting operators leave in-session instructions and the agent acknowledge and act on them. After invoking this skill, the agent knows how to use comment-* tools to:
  • Read comment threads attached to a session or Proof Doc
  • Post new comments to notify operators of progress or findings
  • Respond to operator comments and signal that an instruction has been received
Pair /subtext:comments with live-signal to build a responsive automation loop: poll for new operator signals with live-signal, then use comment-* tools to acknowledge instructions and report back.