Content
70%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 delivers exceptional operational specificity — exact commands, validated JSON skeletons, error-code-driven recovery — and a rigorously gated 9-step workflow with strong validation and idempotency barriers. Its weaknesses are token efficiency (pervasive boilerplate repetition and tangled branching prose that could be decision tables) and underused progressive disclosure, since the 1500-line monolith duplicates material already in the bundled reference file and links to it only as a maintenance note.
Suggestions
Factor the repeated `cd -- '<working_dir>' || { echo "BLOCKED…" }` guard into the one-time script invocation contract (already labeled 'read this once') and state that all subsequent commands implicitly run it, removing ~30 duplicated preambles.
Move the column-type reference table and lookup/M:N body-skeleton material that duplicates references/dataverse-reference.md into that file, and replace the maintenance-note mention with explicit 'See [dataverse-reference.md] for X' navigation links.
Compress the legacy/scoped/fast-path branching in Steps 2–2b into a short decision table keyed on <operation_manifest_mode> values (absent / fallback / candidate / valid), and drop or relocate the time-stamped anecdote ('Empirically verified 2026-05-18…') to a deprecated/notes section.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~1575-line body repeats the `cd -- '<working_dir>' || { echo "BLOCKED…" }` guard in roughly 30 code blocks despite declaring a "read this once" script invocation contract, and contains convoluted multi-paragraph legacy/scoped/fast-path branching plus a time-stamped anecdote ("Empirically verified 2026-05-18 in the chanel-rm demo"). This matches 'noticeably verbose; several unnecessary explanations or padded sections' — not 3 because the padding is pervasive rather than occasional, and not 1 because almost all content is domain-specific operational knowledge rather than generic concepts Claude already knows. | 2 / 5 |
Actionability | Fully executable guidance throughout: exact bash invocations with argument contracts, copy-verbatim JSON body skeletons with explicit "replace only the placeholders in <> brackets" rules, error-code tables (e.g., 0x80044363, 0x80060890), and a concrete auto-rename probe sequence. This matches the 'copy-paste ready code or commands' top anchor. | 5 / 5 |
Workflow Clarity | A numbered 9-step workflow with 'Print before starting' markers, telemetry checkpoints, hard validation gates (decide-before-write barrier, idempotency criterion requiring zero writes on re-run), and explicit error-recovery feedback loops (POST-time collision rescue, cache-invalidation recovery, partial-create recovery). This matches the top anchor of clear sequencing with explicit validation and feedback loops. | 5 / 5 |
Progressive Disclosure | The single bundled reference (references/dataverse-reference.md, well-organized with 15+ sections) is mentioned only once as a maintenance rule ("Keep references/dataverse-reference.md aligned with this skill") rather than as navigation, while the body inlines a column-type table and lookup/alternate-key material the reference already covers; many links target files outside this bundle (../../shared/…, ../add-native/…). This matches 'some structure; references present but not clearly signaled; content that should be separate is inline' — not 4 because the one real reference file is never used as a disclosure target. | 3 / 5 |
Total | 15 / 20 Passed |