Cue plugin
The skill catalog, packaged so coding agents can install it as a plugin rather
than copying markdown by hand. Procedures are the same files as skills/; this
directory is the installable unit around them.
Two MCP servers are bundled. cue answers point lookups —
explain_rule, resolve_owner, resolve_locator, resolve_route,
resolve_api, get_evidence, get_freshness and find_knowledge — against this
project's rules and knowledge base, and writes nothing. Playwright MCP handles
point-in-time browser calls. Exploration still goes through playwright-cli:
an accessibility tree costs more context than the question it answers. See the
ownership table in AGENTS.md.
Install
| Agent | Command |
|---|---|
| Claude Code | /plugin marketplace add wiluszdamian/cue then /plugin install cue@cue |
| Cursor | add wiluszdamian/cue as a marketplace, install cue |
| Codex | codex plugin marketplace add wiluszdamian/cue |
| Grok | grok plugin marketplace add wiluszdamian/cue then grok plugin install cue --trust |
| Gemini CLI | gemini extensions install https://github.com/wiluszdamian/cue.git |
| OpenCode | npx skills add wiluszdamian/cue -a opencode |
| .agents (universal) | npx skills add wiluszdamian/cue -a universal |
OpenCode and the .agents/skills convention do not have a marketplace of their
own that this package can publish to. npx skills add copies the catalog into
the directory those agents already read.
This is the light channel: conventions, not the ESLint guarantee. For enforcement:
npx @wiluszdamian/cue init
Skills
| Invocation | Type | What it does |
|---|---|---|
/cue | model | Use the Cue plugin to write Playwright tests in this repository. |
/compass | user | Name the right Cue skill for the situation and stop. |
/bind | user | Wire this repository to Cue once: check the environment, choose agent targets, and install the constitution and scaffold. |
/survey | user | Map the running application into .agent-kb/app-map by exploring it with playwright-cli. |
/extract | user | Read the product source and write its structure into .agent-kb/product — routes, endpoints, entities, roles, test ids. |
/pin | user | Turn a testing intent into a concrete set of test cases with data, tags and locator candidates. |
/compose | user | Write or change a Playwright test in this repository. |
/inspect | user | Audit an existing Playwright suite against the constitution, and report drift in the knowledge base and routes with no test. |
/resolve-owner | model | Decide which source of Playwright advice governs a topic when two of them disagree. |
/resolve-locator | model | Find the real selector for an element from the project knowledge base, with its freshness and confidence. |
/cue is the on-ramp a model may invoke. The rest of the getting-started
and main-flow skills stay user-invoked so a mention of testing does not start a
survey.
Credits
Cue composes work maintained by others — the official Playwright skills and CLI, and Playwright MCP (Microsoft, Apache-2.0), and a Playwright best-practices reference skill (Currents Software Inc., MIT). It is not affiliated with Microsoft, Anthropic, OpenAI, Google, xAI, or Currents.