--- name: rig-template-write-critique description: A ready-made two-step Rokha Rig pattern — step 1 writes a piece (any writing skill from the live registry), step 2 critiques it against the actual output. Use when the user wants to "write and review", "draft then critique", or wants a worked example of chaining two skills into one workflow. Runs for any agent over Rokha's MCP endpoint, or for a person in the Lab. license: MIT compatibility: Any MCP client that reaches https://rokha.ai/mcp/jsonrpc. Reading the template and searching the registry need no account; building and running need a signed-in account. metadata: author: rokha version: "0.2.0" rokha_kind: rig-template homepage: https://rokha.ai source_repo: https://github.com/rokha-ai/rokha-sdk --- # Rig Template — Write & Critique A two-step workflow skeleton: **create → review**. Step 1 produces a piece of writing; step 2 receives step 1's *actual output* (Rokha threads each step's result into the next) and critiques it with a grade. This is a **template, not a pinned toolchain**: each step carries a registry *query*, and instantiation resolves it against the live Rokha Registry — so the pattern re-shapes around whatever the registry offers today. ## The pattern Read the machine-readable skeleton from `assets/rig.json` (served at `/api/skills/rig-template-write-critique/assets/rig.json`, no auth): - **Step 1 · create** — a writing skill (registry query: `writing`), with a default instruction to produce the requested piece. - **Step 2 · review** — a critique/review skill (registry query: `review critique feedback`), instructed to critique the previous step's output and grade it. ## How an agent runs this 1. Fetch the skeleton: `GET /api/skills/rig-template-write-critique/assets/rig.json` (no auth). 2. For each step, pick a real skill with `registry_search {query: }` over MCP (no auth); `registry_get_skill` reads its full SKILL.md. 3. Sign in (`auth_wallet_challenge` → `auth_wallet_verify`) and build the rig with `rig_author {nodes}`, one node per step with its instruction. 4. `rig_run {rig_id, input}`, then `trace_get {id}`. Each step's output threads into the next, and every step writes a trace. Signed in, 5 runs a day are free; model steps run on your own key. ## In the Lab A person builds and runs the same rig at https://rokha.ai/lab.