Autonomous iterative experimentation loop for any programming task. Guides the user through defining goals, measurable metrics, and scope constraints, then runs an autonomous loop of code changes, testing, measuring, and keeping/discarding results. Inspired by Karpathy's autoresearch. USE FOR: autonomous improvement…
Use when creating, refreshing, expanding, or reviewing a SKRAFT Vally skill evaluation at tests/skills/ /eval.yaml in the skraft-plugin repository. Covers behavior coverage, baseline-versus-isolated-treatment discrimination, natural prompts, outcome rubrics, non-activation cases, regression guards, fixtures, static…
Use this skill BEFORE drafting any agentic primitive module (skill, persona scoping file, scope-attached rule file, orchestrator workflow) or when refactoring an existing one. Activate whenever the task asks to design, restructure, or critique an agentic module across any agent harness (Claude Code, Copilot, Cursor…
Use this agent to design or critique agentic primitive modules (skills, persona scoping files, scope-attached rule files, orchestrator workflows). Activate BEFORE drafting any natural-language primitive content, when refactoring existing modules, or when assessing whether a primitive change adheres to PROSE, Agent…
Reviewer lens: verifies every citation on a handbook page resolves to citations.yml (author, work, year), the referencetemplate required blocks are present on reference pages, and craft vocabulary matches the source skill verbatim. Backed by the deterministic check-citations.mjs tool.
You are an internal worker dispatched by skraft-docs-orchestrator. You make ONE type: derived page (and its FR/EN mirror) faithfully reflect its plugin source.
Reviewer lens: reads handbook pages with one question — is each page written in exactly the Diátaxis mode of the menu section it lives in (tutorial | how-to | explanation | reference), with no mode mixing?
You are an internal worker dispatched by skraft-docs-orchestrator. You write ONE complete type: editorial page (and its FR/EN mirror) as finished, readable prose.
Use to reconcile the SKRAFT handbook (docs/site/) with the shipped plugin sources — detect documentation drift (missing/empty/stale pages, FR/EN parity breaks, broken multi-level menu, ordering gaps, orphan agents/skills/lenses, invalid citations) and repair it through a pull request. Dispatched by the gh-aw docs…
Reviewer lens: reads a handbook page pair with one question — do the FR and EN versions mirror each other (same English basename, same heading structure, equivalent content, neither side ahead)?
You are an internal worker dispatched by skraft-docs-orchestrator. You answer ONE question per item: where does this belong in the book, and in what order? Your only output is a patch to docs/site/data/book.yml.
Reviewer lens: verifies the handbook's multi-level menu is well-formed — Diátaxis mode per part, ordered sections and pages, individual reference items present, and no internal link pointing to a page that does not exist. Backed by the deterministic lint-nav.mjs tool.
Use when adding or changing an agent, sub-agent (worker), reviewer, skill or review lens under plugins/ and the SKRAFT handbook (docs/site/) must reflect the new orchestration chain. Enforces where each agent-chain must be surfaced, FR/EN parity, no dangling derived links, and the validation gate. Load before editing…
Use when writing or editing SKRAFT handbook pages under docs/site/ (FR or EN). Enforces the Diátaxis structure, the Starbucks fil rouge, the artifact-flow connectors between phases, and FR/EN parity. Load before authoring any handbook page, phase page, catalogue page, or deep-dive.
The DISTILL reviewer verifies that scenarios cover acceptance criteria without modifying the .feature files. CQS applies: reading a scenario to evaluate it must not transform it.
BDD scenarios are executable specifications. They capture expected behaviour in a language shared between developers and domain experts, eliminating the ambiguity of textual specifications.
The reviewer is a read-only agent. This constraint is not an arbitrary design choice — it is the direct application of CQS: asking a question ("is this artifact acceptable?") must not change the answer.