Stats
Actions
Tags
From Agency
Use when authoring a new agency capability or extending an existing one — before writing the file. The discipline runs scaffold then lint behind a hard gate; lint must pass before commit.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agency:authoring-capabilitiesThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Three steps; lint gate before commit.
Three steps; lint gate before commit.
Read docs/vision/CAPABILITY-AUTHORING.md
— the unified doctrine. T1 summary: every verb has Spec-016-Hint-#7
markers (Inputs: / Returns: / chain_next:), role-tagged
act / transform / effect, first sentence ≤120 chars.
await call_tool("capability_develop_scaffold_capability",
{"name": "my-cap", "kind": "light"})
Kinds: light (≤3 verbs, no ontology data) · medium (owns ontology
fragment) · heavy (folder-form per Hint #1, ≥3 sibling files).
Emits # agency-scaffold: v1 marker on line 1 — plugin.lint_capability
reads it to block-on-violation rather than warn.
await call_tool("capability_plugin_lint_capability", {"name": "my-cap"})
Five rule families: structural · role_tag · render_slice ·
consumer_contract · token_budget. Mode block + ok: False → fix
before commit. Mode warn → legacy grandfathering (no marker).
npx claudepluginhub netzkontrast/agency --plugin agencyCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.