auto-coding
This skill should be used when the user wants to drive the agent-based implement + test loop over the tasks already authored in a4/{task,bug,spike,research}/. Common triggers include: 'auto-coding', 'implement the tasks', 'run the implementation loop', 'kick off the agents', 'coder', 'agent loop'. Two stages: an autonomous loop body (pick → implement → test, up to 3 cycles), then a user-driven post-loop review that classifies failures or confirms UC ship. Works with or without UCs — UC ship-review is conditional on per-task implements: being non-empty.
Repository SourceNeeds Review
auto-scaffold
This skill should be used when the user needs the LLM to perform project scaffolding — directory structure, dependency installation, build configuration, and a minimal running entry point. Common triggers include: 'scaffold the project', 'set up the project', 'bootstrap the project', 'create the dev environment'. Produces repo files only; does not author any wiki page. Test environment setup is owned by /a4:ci-setup, which can be invoked separately at any time.
Repository SourceNeeds Review
spec
"This skill should be used when the user has converged on the shape of an artifact (format, protocol, schema, renderer rule, CLI surface) and wants to commit it as a prescriptive implementation contract. Writes the spec to `a4/spec/<id>-<slug>.md` with proper frontmatter and body, soft-links any related research tasks (a4/research/<id>-<slug>.md) via standard markdown body links and optional `related:` frontmatter entries, optionally records decision rationale in the `## Decision Log` appendum, and nudges affected wiki pages (architecture / context / domain / actors / nfr). Triggers: 'document this format', 'write up the spec', 'capture this shape', 'this is the spec', 'spec this out', or after the user and LLM converge on a prescriptive shape. Accepts either no argument (extract spec from recent conversation) or a short summary / title (used as a seed). Also handles re-invocation on an existing draft spec to activate it. Requires an `a4/` workspace."
Repository SourceNeeds Review
research
This skill should be used when the user wants to author a single research task — a time-boxed written investigation of a technical topic or comparison of alternatives. Common triggers: 'research X', 'investigate X', 'compare X vs Y', 'evaluate alternatives for', 'I need a research task on'. Required: mode (comparative | single); options (option names) when comparative. The body itself is the deliverable — sources consulted, findings, options. No production code is produced. Writes a4/research/<id>-<slug>.md. `implements:` / `spec:` / `cycle:` are forbidden on research. For exploratory PoC code rather than written investigation, use /a4:spike instead. To discard a task use /a4:discard.
Repository SourceNeeds Review