Content
50%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 presents a clear, well-sequenced six-step workflow with concrete identity-lock guardrails, but it is weakened by verbatim repetition of the identity rules across three sections, an execution example whose flags don't match the actual script, and the absence of any output-validation step for a batch generation operation.
Suggestions
State the identity-lock rules once in a single section and reference it from the workflow steps, removing the near-verbatim duplicates in Step 4 and Guardrails.
Fix the Step 6 command to use only flags scripts/generate-pack.sh actually supports (--image or --image-url, --category, --model, --num, --view, --json, --likeness) and document those options in the body.
Add a validation checkpoint after generation — e.g., instruct the agent to inspect each output image for identity drift and regenerate failures — so the batch workflow includes a feedback loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient domain-specific guidance with no explanation of concepts Claude already knows, but the identity rules are repeated nearly verbatim in three places ("Preserve the exact facial identity from the reference image", "Identity must remain identical to the reference photo" appear in Core Principles, Step 4, and Guardrails). This fits anchor 3 (mostly efficient but could be tightened) better than 2, since the padding is redundancy of novel guidance rather than unnecessary explanation. | 3 / 5 |
Actionability | There is concrete guidance (a bash example, a negative-prompt list, a scene example, a category table), but the Step 6 example passes an "--identity-lock true" flag that scripts/generate-pack.sh does not support (unknown flags are silently discarded), and the script's real interface (--image-url, --model, --view, --json, --likeness) is undocumented in the body. Anchor 3 (concrete but incomplete, missing key details) fits; 4 would require the example to be fully aligned with the executable script. | 3 / 5 |
Workflow Clarity | Steps 1-6 are clearly sequenced (grounding, vision analysis, category selection, prompt construction, negative prompt, execution), but this is a batch operation ("--num 5" image pack) with no validation or verification checkpoint on the outputs — e.g., checking identity fidelity or handling failed generations. Per the judging guidelines, a batch skill without validation is capped at 3 even though the sequence itself is well laid out. | 3 / 5 |
Progressive Disclosure | Sections are organized and the referenced bundle file (scripts/generate-pack.sh) exists, but the body does not surface the script's usage surface (its supported options and defaults) and category prompt templates are embedded inside the script rather than clearly signposted from SKILL.md. Anchor 3 (some structure, references present but not clearly signaled) fits better than 4. | 3 / 5 |
Total | 12 / 20 Passed |