Content
68%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 lean, actionable catalog with concrete probe patterns and an executable YAML config, but it lacks an explicit end-to-end testing workflow with validation checkpoints. Adding a sequenced probe-run-then-verify process would lift workflow clarity.
Suggestions
Add a numbered probe workflow (configure promptfoo YAML -> run -> inspect failures -> iterate top candidates) with explicit validation gates.
Move the per-technique variant lists (e.g., DAN/STAN/AIM personas, encoding schemes) into a references file and link to it, keeping SKILL.md as an overview.
Include a brief "verify bypass" step defining what counts as a confirmed bypass before reporting severity.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Technique entries are terse and assume Claude's knowledge (e.g., one-line Brainfuck/Whitespace examples); only minor phrasing could be trimmed further. | 4 / 5 |
Actionability | Provides concrete encoded-payload examples (Base64, ROT13, leet, morse) and a copy-paste promptfoo YAML probe block, with only minor gaps in coverage. | 4 / 5 |
Workflow Clarity | The content is a technique catalog rather than a sequenced workflow; the only checkpoint is "reproducible across 3+ fresh conversations," and batch probe runs lack explicit validate-then-fix loops. | 3 / 5 |
Progressive Disclosure | No bundle files exist; the single SKILL.md is well-organized into Techniques, Probe pattern, Detection, Severity, Defender, and Cross-references sections with a single one-level external reference. | 4 / 5 |
Total | 15 / 20 Passed |