A single anonymized seat on the Elenchus council — a thin sandbox the chairman dispatches in parallel, one per model tier (Fable/Opus/Sonnet/Haiku). Its persona, round, and exact output schema arrive INSIDE the chairman's dispatch prompt (composed from the engine's seat-base + tier adapter + the front end's round…
Gives the agent up-to-date documentation and code examples for libraries and frameworks, fetched on demand so answers are not stuck at the model's training cutoff. Runs locally from the @upstash/context7-mcp npm package. Needs 1 environment variable to run.
Lets the agent drive a real browser: open pages, click, type, take screenshots and read the accessibility tree, using Playwright. Runs locally from the @playwright/mcp npm package.
Claude Code instructions for alexvtejeda/elenchus, covering claude.md, what elenchus is, durable state (checkpoints), mcp servers (context7 + playwright) and conventions.
A single browser-verification seat for the finishing-implementation-elenchus skill — a thin, report-only sandbox the chairman dispatches (one at a time, pinned to a model tier) to drive the running app via Playwright MCP and report which acceptance-checklist items PASS / FAIL / need a human / are MISSING. It NEVER…
A single terminal-render verification seat for the elenchus-tui skill — a thin, report-only sandbox the chairman dispatches (one per breakpoint, in parallel, pinned to a model tier) to drive a pure-ASCII TUI mockup inside its OWN isolated tmux session and report whether its assigned breakpoint renders cleanly: PASSED…
Use when the user wants a build or architecture idea stress-tested before they commit to building it — "help me plan/design this app", "design the architecture for X", "review my architecture before I build", "is my mental model of X sound", "I want to build Y because Z — poke holes in it". For the still-articulating…
Use when a request asks to convene the council, stress-test a premise, or get a multi-seat anonymized critique before committing to a design — or when an Elenchus front-end skill (e.g. elenchus-build) needs to run that review. This is the shared engine: front ends call it; it is not triggered directly by ordinary…
Use when the user wants to build a CLOSED CORPUS of real, verified resources or links by fanning out parallel agents — "gather as many X links as possible", "harvest a curated set of sources", "build a closed space of playlists/repos/feeds for a model (or person) to choose from", "collect verified URLs across these…
Use when you have an approved spec (or firm requirements) for a multi-step build and want the implementation plan itself stress-tested before you execute it — "write the plan and have the council check it", "verify this plan against the spec", "turn this spec into a plan I can trust", or the writing-plans stage of the…
Use when the user wants to research or learn a topic before forming a firm opinion — "I want to learn about X", "help me research Y", "gather a reading list / resources / a study path on Z", "I'm new to X and want to understand it deeply", "convene a research council on X". The research/study front end over the…
Use when the user wants a terminal-UI (TUI) mockup designed and then proven to actually render — pure-ASCII, no framework — before building the real thing. Triggers: "mock up a terminal UI for X", "design a TUI / CLI dashboard I can show someone", "make an ASCII wireframe of this terminal app and make sure it…
Use when a frontend feature has been implemented and you want to verify it actually works in the running browser against its spec/plan — then fix what doesn't — before calling it done. Triggers: "verify my implementation works", "check the frontend against the spec", "did I actually build everything", "finish and…
Use to SHOW the user something in their browser during an Elenchus or brainstorming flow — UI mockups, wireframes, architecture diagrams, side-by-side visual comparisons, design polish — when the content itself is visual and the user would grasp it better by seeing than by reading. The user always ANSWERS in the…