Split out of #563. plugins/ca/commands/release.md currently documents three arguments the skill it routes to has never implemented:
argument-hint: ["<version>"] | --auto | --dry-run
The skill's only argument is the target plugin. --auto, --dry-run, and an explicit <version> appear nowhere in plugins/ca/skills/release/SKILL.md — they are fossils of the portable pre-#125 version of the skill.
#563 reconciles the command surface to the skill, which means deleting those three from the documentation. That is correct as documentation hygiene, but --dry-run in particular describes something genuinely useful that a user may have tried and assumed was broken rather than absent: run every gate, surface the readiness report, compose no tag.
This issue tracks implementing it for real rather than letting the capability disappear silently.
Shape
The skill already has a natural stopping point. Pre-flight plus Phase 1 derive the version, classify every commit in the payload-scoped window, and validate the changelog footers — all read-only. Phase 2 is where the annotated tag is composed. A dry run is therefore "execute through Phase 1, emit the report, STOP before composing."
Worth deciding at implementation time, not now:
- whether pre-tag check commands run in a dry run (they are non-mutating by DECISION-0034, so probably yes)
- whether it reports the resolved row for the target, which would double as a way to validate a freshly authored
.codearbiter/release-targets.md without tagging anything
- whether it belongs as a flag or as a separate read-only command
Dependency
Blocked on #563 landing. Implementing a dry run against the current hardcoded lane would have to be rewritten immediately for the declared-file lane.
Split out of #563.
plugins/ca/commands/release.mdcurrently documents three arguments the skill it routes to has never implemented:The skill's only argument is the target plugin.
--auto,--dry-run, and an explicit<version>appear nowhere inplugins/ca/skills/release/SKILL.md— they are fossils of the portable pre-#125 version of the skill.#563 reconciles the command surface to the skill, which means deleting those three from the documentation. That is correct as documentation hygiene, but
--dry-runin particular describes something genuinely useful that a user may have tried and assumed was broken rather than absent: run every gate, surface the readiness report, compose no tag.This issue tracks implementing it for real rather than letting the capability disappear silently.
Shape
The skill already has a natural stopping point. Pre-flight plus Phase 1 derive the version, classify every commit in the payload-scoped window, and validate the changelog footers — all read-only. Phase 2 is where the annotated tag is composed. A dry run is therefore "execute through Phase 1, emit the report, STOP before composing."
Worth deciding at implementation time, not now:
.codearbiter/release-targets.mdwithout tagging anythingDependency
Blocked on #563 landing. Implementing a dry run against the current hardcoded lane would have to be rewritten immediately for the declared-file lane.