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.
The skill delivers an exceptionally rigorous, well-sequenced lifecycle workflow with strong validation and rollback discipline, and its guidance is mostly concrete thanks to real CLI commands and exact field names. Its weaknesses are structural: everything is inlined in one ~293-line file with dense spec prose (especially the rerank rules) that belongs in reference files, and a few operational steps lack executable commands.
Suggestions
Move the three ranking-quality rules in Section 6 (stratification tiers, VOI cost divisors, load-bearing audit, orthogonal cap defaults) into a references/ranking-rules.md file and keep only a short summary plus the CLI commands in SKILL.md.
Show the exact invocation of plan_material_single_move (as is already done for plan-wave-rerank) so the single-move path is copy-paste executable like the rest of the workflow.
Trim the introductory enumeration of all host surfaces and skills directories to the one example plus a one-line pointer to 'loopx project-skill install --help' to cut token cost without losing actionability.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body never explains concepts Claude already knows and assumes competence throughout, but Section 6's nested ranking-quality rules and the intro's enumeration of every host surface and skills directory are over-long spec prose that could be trimmed. It sits between 'minor instances of over-explanation' (4) and 'every token earns its place' (5), closer to 4. | 4 / 5 |
Actionability | Concrete commands are present ('loopx project-skill install --project . --skill loopx-material --surface codex --execute', 'loopx material-lifecycle architecture --format json', 'loopx material-lifecycle plan-wave-rerank --input <path-or-->') along with exact field names, defaults, and dispositions (top_window/ranked_backlog/no_change, max_orthogonal_in_top_window default 0). Minor gaps remain: the invocation shape of plan_material_single_move is never shown, and steps like 'record source revision and read scope' name no command. | 4 / 5 |
Workflow Clarity | Eight numbered workflow steps with explicit validation throughout: snapshot requires a verified backup, Step 4 requires repair-or-rollback before reporting intake complete, Step 8 is an apply-gate checklist (digests match, parse errors zero, counts exact, CAS revision, rollback rehearsed, owner gate) with post-apply readback and mismatch handling, plus a dedicated Stop Conditions section. This matches the anchor for clear sequence with explicit validation, feedback loops, and checklists. | 5 / 5 |
Progressive Disclosure | The single 293-line SKILL.md has no bundle files and no references at all; spec-detail content that belongs in separate files is inlined, including the three dense ranking-quality rules in Section 6, the CLI surface/directory enumeration, and the adapter contract details. Internal section organization is good, but the rubric's intent that SKILL.md be an overview pointing to detailed materials is not met. | 3 / 5 |
Total | 16 / 20 Passed |