A reusable skill definition — a SKILL.md file (plus any supporting reference files, scripts, or prompt templates in its directory) that captures a proven technique, pattern, or reference so future agents can discover and apply it. Written by sp-writing-skills via its TDD-for-documentation loop: YAML frontmatter with name + a when-to-use-only description, then a body optimized for discovery (keyword coverage, cross-references, flowcharts only where a decision is non-obvious).
What makes it distinctive is that it is executable process knowledge: unlike a artifact-solution-doc (a lesson agents read as grounding) or an artifact-adr (a one-time decision record), a skill doc changes what the agent can do — once in the library it triggers automatically (via sp-using-superpowers) and is followed as a gated procedure. It is the unit that makes the superpowers library grow itself, the capability- compounding output of stage-learn.
Produced by (backlinks)
- sp-writing-skills — authors/edits a skill, test-first against subagent pressure scenarios.
- gstack-skillify — the gstack counterpart: codifies a successful run into a permanent browser-skill (same capability-compounding idea, a scrape-specific instance).
pstack:
- pstack-create-verification-skill — a project-local
verify-<app>skill with Launch / Doctor / Drive / Evidence / Cleanup / Helpers sections, written for an agent that will read it cold mid-task. - pstack-automate-me — a personal
<handle>-modeskill drafted from the user’s own transcripts, updated in place on repeat runs. - mp-writing-for-agents — Matt Pocock’s model-invoked reference for writing any agent-facing document (skills,
AGENTS.md/CLAUDE.md, pointed-at docs); fires whenever one is created or edited (v1.2.0). - pstack-reflect — targeted edits to existing skills, each routed from a synthesized learning and gated by a structural-enforcement check.
- pstack-maintain-verification-skill — proven corrections to a generated verification skill and its feature map.
Tessl — the layer that treats this artifact as a tradeable dependency rather than as a note to self:
- tessl-skill — scaffolds, lints, publishes, and unpublishes it;
skill importpromotes a looseSKILL.mdinto an artifact-capability-bundle, which is the prerequisite for evaluating it at all. - tessl-review-fix — rewrites it iteratively until it clears a quality threshold.
A registry changes what the artifact is: it acquires a version, an owner, a visibility setting, a security score, an eval suite, and a rollback path. See artifact-capability-bundle for the shape it ships as.
See Also
- artifact-capability-bundle — the versioned package a skill ships inside once it is distributed rather than written in place.
- artifact-eval-suite — the scenarios that prove a skill changes behavior.
- artifact-solution-doc — machine-consumable lessons agents read (vs a skill agents run).
- artifact-explainer · artifact-retrospective — the other stage-learn outputs (human-facing knowledge).
- pattern-knowledge-compounding — the technique this artifact realizes (capability-compounding form).
- stage-learn — the stage this artifact is an output of.