Content
81%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 highly actionable with executable code, well-sequenced workflows including validation and rollback, and a clean progressive-disclosure structure backed by real bundle files. Its main weakness is token efficiency: overlapping model-strategy tables, duplicated inline templates, and time-sensitive test narrative inflate the body beyond overview size.
Suggestions
Collapse the three overlapping model-strategy mappings (Overview, Step 0 strategy table, Model-Specific Notes) into a single table.
Remove the full system-prompt templates inlined in Steps 2 and 6 and point to references/jailbreak-templates.md, which already contains them.
Move the 'Tested Results (March 2026)' section and the trigger-word list into a reference file (or a clearly marked dated/deprecated section) to keep the body lean and age-resistant.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Most content is skill-specific and non-generic, but the ~400-line body includes redundancy (model-strategy mappings appear in three places: the Step 0 strategy table, the Model-Specific Notes table, and the Overview), a 34-word inline trigger-word list that belongs in references, and a time-sensitive 'Tested Results (March 2026)' narrative with model-version specifics that is not placed in a deprecated/old-patterns section. Mostly efficient but could be tightened matches anchor 3. | 3 / 5 |
Actionability | The body provides copy-paste-ready, executable guidance throughout: the load_godmode.py exec snippet, auto_jailbreak() calls with arguments, concrete config.yaml and prefill.json contents, parseltongue CLI one-liners, race_models() invocation, and a complete end-to-end pipeline example. Referenced script paths were verified to exist in the bundle. | 5 / 5 |
Workflow Clarity | Steps 0-6 are explicitly sequenced with a decision table for mode selection, and validation/feedback loops are present throughout: baseline refusal testing before jailbreaking, dry_run=True to test without writing config, undo_jailbreak() for rollback, a canary query with scoring thresholds, and a clear escalation ladder (system prompt -> prefill -> Parseltongue -> encoding escalation -> multi-model racing). | 5 / 5 |
Progressive Disclosure | References are clearly signaled ('See references/jailbreak-templates.md for all templates', 'See scripts/parseltongue.py for the Python implementation') and all six referenced bundle files exist, one level deep. However, the full jailbreak system-prompt template is inlined in both Step 2 and Step 6 despite references/jailbreak-templates.md existing precisely for that content, and the trigger-word list and test-results narrative are inlined rather than split out — good structure with minor organization gaps matches anchor 4. | 4 / 5 |
Total | 17 / 20 Passed |