Content
47%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is maximally lean but is pure delegation: it directs Claude to read and execute an external playbook that is absent from this bundle (and, given the skill's own name, possibly a circular self-reference), then follow additional missing rule files. No task steps, validation checkpoints, or navigable in-bundle content exist, leaving the skill unexecutable as written.
Suggestions
Inline the core publish workflow (format conversion, approval gate, publishing steps) or bundle the authoritative playbook as a references/ file instead of pointing to `.claude/skills/publish-to-confluence/SKILL.md`, which is absent from this bundle and plausibly circular given this skill's own name.
Add explicit validation and approval-gate checkpoints in the body — publishing to Confluence is an outward-facing operation, so the absence of verification steps caps workflow clarity.
Make the Antigravity substitution rule concrete with a worked example mapping one `.claude/agents/<name>` path to its `.agents/agents/<name>.md` counterpart.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is five lines with zero padding or over-explanation — every sentence is an instruction or a substitution rule. This matches the level-5 anchor (lean and efficient, every token earns its place). | 5 / 5 |
Actionability | The pointers are concrete (exact paths, the "invoke_subagent" tool name), but the specific steps to execute the actual publishing task are entirely delegated to files that do not exist in this bundle — and the primary target, `.claude/skills/publish-to-confluence/SKILL.md`, is plausibly a circular self-reference since this skill is itself named publish-to-confluence. Nothing executable for the task itself is provided, matching the level-2 anchor (minimal concrete guidance, missing the specific steps to execute). | 2 / 5 |
Workflow Clarity | Only a rough meta-sequence exists (read playbook → execute it as written → follow cog.md rules), with no task-level steps and no validation or approval-gate checkpoints in the body for an outward-facing publish operation. This matches the level-2 anchor (rough sequence present but many gaps, validation absent) and falls below the level-3 cap that applies to operations like publishing. | 2 / 5 |
Progressive Disclosure | Scored against the actual bundle structure: no references/, scripts/, or assets/ directories exist, and all three referenced paths (`.claude/skills/publish-to-confluence/SKILL.md`, `.agents/rules/cog.md`, `.agents/agents/<name>.md`) live outside the bundle. The reference chain also nests multiple levels (SKILL.md → playbook → agent workers → substituted agent files), matching the level-2 anchor (minimal structure, references buried/unavailable). | 2 / 5 |
Total | 11 / 20 Passed |