@subtext
Tools that let coding agents verify and show their work against your running application
Comment MCP tools for agent-user collaboration. Use when reviewing sessions or live pages to leave observations, read user feedback, reply, and resolve.
Proof document MCP tools for creating, updating, and closing agent work documentation. Use when tracking a bug fix, UX review, or changeset to produce a permanent, evidence-backed record.
Agent explores the user's site via hosted broswer (live), leaving comments as it goes. Accepts a user-described flow or explores organically. Capped at ~10 interactions across 2-3 pages. Returns session URL, trace URL, and metrics.
Live browser MCP tools for driving a hosted browser — connections, views, interactions, console, network, and tunnel. Use when reproducing flows, taking screenshots, or interacting with a running app.
Interactive first-run onboarding for new Subtext users. Connects to the user's local dev server, proves a small change with before/after evidence in a watchable trace, then bootstraps a starter sightmap from what was learned.
You MUST use this skill when implementing, fixing, or refactoring code. Captures evidence artifacts (screenshots, network traces, code diffs, trace session links) into a proof document as you work.
Short recipe to create sightmap definitions for a project from scratch.
Review a completed Subtext session and produce a structured summary. Use when you have a session URL and want to understand what happened — whether to verify another agent's proof work, walk through a dev / staging / preview flow, or summarize a captured session. Optionally emits reproduction steps on request (execution lives in `subtext:live`). Skip for tasks that modify code (use `subtext:proof`) or drive a running app (use `subtext:live`).
Session replay tools for analyzing Fullstory session recordings. Sparse API catalog — tools are self-describing.
Install the Subtext plugin and configure MCP servers. Authenticates via OAuth or API Key.
Foundation skill for the subtext plugin. MCP tool conventions, environment detection, security rules, and sightmap upload.
Use when setting up the sight map (.sightmap/ YAML files) — defining components, views, requests, or other runtime semantics for the application. Also use when snapshot output shows generic a11y roles instead of meaningful names.
Use when opening a hosted browser connection against a localhost or local dev server URL. Sets up a reverse tunnel so the hosted browser can reach the user's local server.
Use when starting any conversation that may involve rendered UI, observed sessions, or producing reviewer-facing evidence (screenshots, viewer links, code diffs, command output). Establishes how subtext skills compose and when to invoke them before any response or action.