Content
77%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.
A highly actionable, well-sequenced orchestration skill with concrete invocations, constants, checkpoints, and templates throughout. Its weaknesses are token efficiency (systematic duplication of the robotics frame and sim-first/hardware rules across sections) and structure (a monolithic single-file layout with no bundle split despite referencing shared files).
Suggestions
Collapse the Overview's constraint bullet list into the Phase 0 'Robotics Problem Frame' so embodiment/task/interface fields are stated once, and cut the repetition of the sim-first/no-hardware rule from Constants, Execution Rule, and Phase 3 down to a single authoritative statement in Key Rules.
Split the full Phase 6 report skeleton and the Phase 1 landscape-matrix table into a reference file (e.g. references/report-template.md) and link to it, keeping SKILL.md as an overview.
Trim the verbatim checkpoint message templates to short bullet cues (idea, embodiment, benchmark, risk) instead of full emoji-formatted message bodies, since Claude can compose the prose itself.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body does not explain concepts Claude already knows (RL, BC, sim2real are used tersely), but it is systematically redundant: the Overview's constraint axes (embodiment, task family, observation/action interface, sim-first default) are restated nearly verbatim in the Phase 0 frame, and the sim-first/no-hardware rule is repeated in Constants, Execution Rule, the Real Robot Rule, Key Rules, and Phase 3. Checkpoint message templates add further padding that could be tightened. | 3 / 5 |
Actionability | Every phase gives exact executable invocations (e.g. '/research-lit "$ARGUMENTS — focus venues: CoRL, RSS, ICRA, IROS, RA-L, TRO, Science Robotics"'), concrete constants (MAX_PILOT_IDEAS = 3, PILOT_MODE = sim-first), a copy-paste pilot-design field list, a full report skeleton, and verbatim checkpoint messages. This is fully copy-paste ready for an orchestration skill. | 5 / 5 |
Workflow Clarity | Phases 0-6 are clearly sequenced with explicit checkpoints after Phase 1 and Phase 2, decision branches for each possible user response (approve / request changes / narrow scope), AUTO_PROCEED fallback behavior, an explicit rule to continue past Phase 3 even without execution, and a re-run feedback loop ('refine the robotics frame, re-run Phase 1, and present again'). This matches the top anchor including error-recovery loops. | 5 / 5 |
Progressive Disclosure | There are no bundle files (references/, scripts/, assets/ are absent) and the entire 370-line skill is one monolithic document: the full report template, the landscape-matrix guidance, and the pilot-design template are all inline where an overview-plus-references split would fit. The shared-references links (reviewer-backends.md, output-versioning.md, output-manifest.md, output-language.md) are clearly signaled, but the dense reviewer-routing blockquote at the top largely duplicates what the linked shared reference governs. | 3 / 5 |
Total | 16 / 20 Passed |