Skip to content

fix: declare the release skill's argument the way skills take arguments - #112

Merged
Raphael-Hao merged 1 commit into
mainfrom
fix/release-skill-named-arg
Aug 25, 2026
Merged

Raphael-Hao merged 1 commit into
mainfrom
fix/release-skill-named-arg

Conversation

@Raphael-Hao

Copy link
Copy Markdown
Collaborator

Follow-up to #111. /release minor showed no argument hint, because the argument was declared the wrong way.

What was wrong

The bump argument was written as $1 — the shape a shell script or a plugin command takes, not a skill. Claude Code's bundled skill-authoring guidance is explicit:

arguments and argument-hint: Only include if the skill takes parameters. Use $name in the body for substitution.

The skill had argument-hint but no arguments, so there was nothing to substitute and nothing to hint with: $1 reached the body as literal text.

What changed

  • arguments: [bump] declared, and the body reads $bump throughout (inputs section, step 2 prose, the shell snippet).
  • when_to_use added. That is the field auto-invocation keys off; the trigger phrases were sitting in description, which should just say what the skill does. Phrases moved, description trimmed.

Verification

Read the frontmatter schema out of the Claude Code build in use here (2.1.245). It accepts exactly argument-hint, arguments, disable-model-invocation, user-invocable, context, agent, paths, settings — so argument-hint was a valid key all along; it was the missing arguments companion that made it inert.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NhcXpAA66z8a1mRJzARJqy

The bump argument was written as `$1`, which is the shape a shell script or a
plugin command takes — not a skill. Claude Code's own skill-authoring guidance
is explicit: `arguments` and `argument-hint` go together, and the body
substitutes `$name`. Without the `arguments` declaration there was nothing to
substitute and nothing to hint with, so `/release minor` offered no completion
and `$1` reached the body as literal text.

Declares `arguments: [bump]` and reads `$bump` throughout. Adds `when_to_use`,
which is the field auto-invocation actually keys off — the description alone
was carrying trigger phrases that belong there, so the phrases move and the
description goes back to saying what the skill does.

Verified against the frontmatter schema in the Claude Code build in use here
(2.1.245), which accepts exactly: argument-hint, arguments,
disable-model-invocation, user-invocable, context, agent, paths, settings.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NhcXpAA66z8a1mRJzARJqy
@Raphael-Hao
Raphael-Hao merged commit 929fe14 into main Aug 25, 2026
5 checks passed
@Raphael-Hao
Raphael-Hao deleted the fix/release-skill-named-arg branch August 25, 2026 05:34
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant