Content
85%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.
Highly actionable, platform-specific guidance with a well-sequenced workflow, explicit error-recovery loops, and concrete correct/incorrect code examples. The main weaknesses are a monolithic 420-line body with no reference-file split, and minor repetition of the slots registration example.
Suggestions
Split detailed per-topic guides (Register Client UI variants, Themes and styles, Versions/approval/repair, Common failure checks) into one-level-deep reference files (e.g., references/client-ui.md, references/versions.md) and keep SKILL.md as the workflow, platform-choice rules, and core code constraints with clearly signaled links.
Deduplicate the three near-identical ctx.get('slots') + slots.inject examples into one canonical registration example, and cite it from the Settings pages / Run-specific panel sections instead of repeating the full snippet.
Consolidate the repeated 'query the Provider before writing code' rule (stated in the intro, workflow, tool guidance, and Provider navigation sections) into a single stated-once constraint.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious platform constraints and assumes JS competence, but the ctx.get('slots') + slots.inject example appears nearly verbatim three times (Execution environment, Register Client UI, Cordis Run-specific panel) and 'query before writing code' is restated across several sections. Efficient overall with minor trimmable repetition — not 5 (redundancy exists), not 3 (nearly all tokens carry platform-specific rules Claude cannot know). | 4 / 5 |
Actionability | Executable, copy-paste-ready JavaScript function bodies with correct/incorrect pairs (e.g., the timer examples with explicit 'The first example does not declare the timer hard dependency'), a run-mode selection table, a tool usage table, and a failure-symptom-to-check table cover the common cases concretely. Placeholders like 'target.slot' are justified because real names must come from runtime inspection. | 5 / 5 |
Workflow Clarity | A numbered 7-step workflow with explicit asynchronous/approval handling ('After cordis_run returns awaiting-approval or starting, end the current Tool flow'), an explicit error-recovery loop (inspect_self diagnostics → re-query Provider → define new Package → run again), and rollback rules. Validation checkpoints and feedback loops are present and explicit, matching the anchor-5 example. | 5 / 5 |
Progressive Disclosure | The single 420-line SKILL.md inlines detailed per-topic guides (Client UI registration variants, themes, versions/approval, failure checks) that belong in separate reference files, and no bundle files exist at all. Good section headers provide structure — so not 2 — but content that should be separate is entirely inline with no one-level-deep references, capping it at 3. | 3 / 5 |
Total | 17 / 20 Passed |