Content
77%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.
A well-organized, highly actionable body with excellent workflow discipline (blast-radius gate, peer review, abort criteria) that clearly distinguishes safe chaos practice from sabotage. The two real defects are the broken bundle — every referenced reference, script, and asset path is missing — and the designer-markdown/postmortem-json format mismatch in the Quick start chain.
Suggestions
Ship the referenced bundle files: create references/chaos_principles.md, references/experiment_design.md, references/attack_taxonomy.md, references/tooling_landscape.md, the three scripts/*.py tools, and the two assets/*.md templates — currently every path the body points to is missing, so the Quick start cannot run.
Fix the format mismatch in the primary workflow: experiment_designer.py 'Outputs a markdown plan' but experiment_postmortem.py consumes '--plan experiment.json' — either have the designer emit JSON or document the expected postmortem input format.
Trim duplication between the Quick start and the '3 Python tools' section (the same three commands appear twice), and cut the Netflix four-principles recap to a one-line pointer to references/chaos_principles.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly dense and useful (attack table, tooling chooser, workflows, anti-patterns), but the Quick start repeats all three tool invocations that the '3 Python tools' section restates with fuller arguments, and the Netflix four-principles recap is largely material Claude already knows. | 4 / 5 |
Actionability | Copy-paste-ready CLI blocks with full flags for all three tools, plus concrete decision rules for tool selection and attack choice. Minor gap: the Quick start chain is inconsistent — the designer 'Outputs a markdown plan' but the postmortem consumes '--plan experiment.json', a file format nothing in the skill produces. | 4 / 5 |
Workflow Clarity | Workflow 1 has explicit validation gates ('Run blast_radius_calculator.py — confirm GREEN before proceeding', peer review that abort criteria are concrete) and an error-recovery path ('If abort criteria are hit, abort immediately; record what happened'), so the destructive-operation cap does not apply. Workflows 2 and 3 are clearly sequenced with progression steps. | 5 / 5 |
Progressive Disclosure | References are clearly signaled and one level deep (a dedicated References section listing 4 files, plus 2 assets and 3 scripts), but none of these files exist in the bundle — no references/, scripts/, or assets/ directories are present, so every referenced path is dead and the Quick start commands cannot run. Structure is good; navigation is broken. | 3 / 5 |
Total | 16 / 20 Passed |