Advertise an existing capability on your A2A agent card so other agents can discover and request it. A card entry does not implement a new handler. To teach your own agent a procedure, write a [`SKILL.md` skill](/guides/skills).

An ADR that doesn't pass the docs build or drifts from the house shape costs a review round-trip. This is the checklist that makes an agent-authored ADR land first try.

The setup wizard already handled the mechanical work — archetype, bundle, persona. This skill is the part that has to happen *in conversation*: proving the setup on the operator's real work and personalizing it. One step at a time, every step skippable, a few sentences per message. If they detour

Use when starting documentation work in an unfamiliar codebase or feature; establish readers, source evidence, content ownership, and validation commands.

Use when improving documentation information architecture, audience paths, navigation, page ownership, or reorganizing sections.

Use after documentation changes to check accuracy, links, generated content, builds, review, and authorized publication or merge.

A schedule runs on the clock; a drop zone runs on arrival. protoAgent's watch primitive polls a condition out-of-band and runs a follow-up prompt in this session when it moves. The tools: `create_watch`, `list_watches`, `clear_watch`; the folder check is `cowork:folder_changed`.

Drive the `due-diligence` workflow (plan M2): a codebase-mapper reads what our code already does about the question while a researcher gathers the external evidence; an antagonist steelmans the opposing case and a verifier checks the load-bearing claims; the synthesis is a verdict document that ends

You are the navigator; the operator drives. **Each numbered step is ONE turn.** Do the step, show the evidence, then **end the turn with the question shown.** Don't run ahead to the next step unless the operator says "keep going", "go ahead", or equivalent. Open each turn with a one-line "where we a

Use when asked to rewrite, simplify, tighten, clarify, or "make readable" an existing text (a doc, email, policy, message, or pasted paragraph), or to translate jargon into plain English. Rewrites it in plain language, keeps every fact, and reports what changed.

The friction ledger is how this system improves itself. Auto-capture already logs tool errors and shell commands that duplicate a tool you have (`cat` when `read_file` is bound); `git`, tests and builds through a shell tool are normal work, not friction. It cannot see the thing only you know — tha

The console has an **Artifact panel** powered by the `show_artifact` tool. Use it whenever the user wants to **look at something rendered** rather than get source files.

protoAgent's scheduler runs full agent turns on a cadence — any cron expression, or a one-shot at an ISO timestamp. The tools: `schedule_task`, `list_schedules`, `cancel_schedule`.

Each agent records its own rough edges into its own ledger. One agent's ledger is a list of complaints; **all of them read together are a map of what the harness gets wrong.** This skill is the pass that reads the map.

Deliverables are files: write the document with `execute_code` using `python-docx`, save it into the operator's fenced project folder (if there are several, ask once which and remember it), and report the saved path. If the library is missing, say so — it's declared in this plugin's `requires_pip`

Learn the operator's writing voice from samples they provide and save it as a reusable style profile, so drafts written in their name sound like them. Use when the operator wants drafts to match their style, offers writing samples, or asks to set up/update their voice profile.