skill-author

State: open/active · Genre: plan

Status: working draft, 2026-10-02. Part of the strategy hub. The first skill in the prose package (Skills).

skill-author teaches how to write a minimal skill: what one is, how to write it, and how to grade it. It is written to its own rules.

What a minimal skill is

One process, one skill. A terse, succinct, clear description of one process, done the way we want it done.
A scope line. What the skill does, in one sentence, and what it does not do.
Inputs and outputs. What it needs to start and what exists when it ends.
Steps. Numbered, short, each one an action. No background essays; link out instead.
Extension points (optional). The lifecycle hooks a project can fill through sdlc/extend/skills/<skill>.lifecycle (Skill lifecycle extensions).
Grading (optional). How to tell a good result from a bad one, in a few checks.
Test environment (optional). How to set up a scratch environment to run the skill against, and what to look at afterwards.

Script-free first. A skill gets a script or an sdlc wrapper only when a step is deterministic and repeated.

Ask for, or read, the process the skill describes, and its one-sentence scope.
Check the catalog for an existing skill that covers it. Narrow or extend that one instead of adding a second.
Write the skill in the minimal shape above, in the vault format: <name>/<name>.md with ap-kind: skill and ap-plugin frontmatter.
Add extension points only for steps a project is likely to change.
Write the grading checks, and the test environment if the skill changes files or calls tools.
Grade the draft against the checks below and tighten it until it passes.
For an sdlc wrapper, run the skill-prose gate and agent-pants build and check.

The scope fits in one sentence, and the out-of-scope line exists.
Every step is an action a reader can take without guessing.
Nothing is explained that a link could cover.
Extension points, if any, are named and each has a stated position.
The grading checks can be answered yes or no.
An agent given only the skill and its inputs produces the expected outputs in the test environment.

Harness-specific packaging. agent-pants renders each harness from the vault.