Content
28%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 a hollow template: boilerplate instructions that would fit any skill, no projection-specific techniques, steps, or examples, and its entire depth delegated to a resource file that is absent from the bundle. Only the Use/Do-not-use trigger lists and the reference structure show real skill design.
Suggestions
Create the referenced `resources/implementation-playbook.md` (or fix the path to an existing bundle directory like `references/`) — currently every substantive detail is deferred to a file that does not exist, breaking progressive disclosure entirely.
Replace the generic Instructions bullets with concrete projection-building guidance: e.g., defining a projection schema from event types, idempotent event handlers keyed on event ID, checkpoint/offset tracking for the event store, and rebuilding/replaying a projection from stream start.
Add a real workflow with validation checkpoints — e.g., verify projection state against a replayed event count, handle versioned events, and reconcile drift between the read model and the event store — instead of the abstract 'validate outcomes' bullet, and remove the tautological 'Do not use' bullets to reclaim tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is short but a large share of its tokens are non-earning generic filler: "Clarify goals, constraints, and required inputs. Apply relevant best practices and validate outcomes" and "Provide actionable steps and verification" could appear verbatim in any skill and teach nothing Claude doesn't know, and the "Do not use this skill when" bullets ("The task is unrelated to projection patterns") are tautological. It is not a 2 because the total length is small and the Use-when list carries real signal; it is not a 4-5 because the Instructions section is pure boilerplate padding. | 3 / 5 |
Actionability | There is no concrete guidance anywhere: no code, no commands, no named techniques or patterns (no event replay, idempotency handling, catch-up subscriptions, projection versioning, checkpointing, etc.), just abstract directives like "Apply relevant best practices and validate outcomes". This matches the 1 anchor ('entirely vague or abstract; only describes rather than instructs'); even the 2 anchor's 'high-level hints' are absent, and the single concrete pointer targets a file that does not exist in the bundle. | 1 / 5 |
Workflow Clarity | The Instructions bullets form only a trivial, task-agnostic meta-sequence (clarify → apply → provide) rather than a projection-building workflow, and validation appears only as the abstract phrase "validate outcomes" with no checkpoint or error-recovery guidance. This is a rough sequence with many gaps matching the 2 anchor; it is not a 3 because there is no projection-specific sequenced process, and not a 1 only because a nominal step order does exist. | 2 / 5 |
Progressive Disclosure | Structurally the pattern is right — a short overview with a single clearly-signaled, one-level-deep reference ("`resources/implementation-playbook.md` for detailed patterns and examples") — but the referenced file does not exist in the bundle (no resources/, references/, scripts/, or assets/ directories present), so the disclosure chain dead-ends and the skill defers 100% of its substance to a missing file. This lands at the 3 anchor ('some structure but could be better organized') — better than 2 because the reference is prominently signaled rather than buried, but the broken target leaves no navigable content. | 3 / 5 |
Total | 9 / 20 Passed |