This repository was archived by the owner on Oct 21, 2025. It is now read-only.
LLM-generated prompts #7
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Implement 3-agent system (exploit generator, target, evaluator) - Support configurable multi-turn attacks (1=single, 2+=trust-building) - Add session persistence and learning capabilities - Track success patterns per attacker model - Create CLI entry point `uv run interactive` - Default to Claude Opus 4.1 as attacker, evaluator defaults to same - Include 8 attack strategies (trust building, side tasks, etc.) - Export sessions to timestamped JSON with full dialogue history 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <[email protected]>
Major improvements: - Split planning and execution stages for more natural attacks - Add `uv run attack` for focused attack runs with 3 modes: * Predefined strategies * Custom goals (describe what to test) * Custom prompts (provide exact prompts) - Add `uv run sessions` for viewing/analyzing past sessions - Never truncate responses in display (show full content) - Support multiple attacker models (Claude, GPT-4, Llama, Gemini, etc.) - Per-model learning system tracks what works for each attacker - Real-time display of conversation stages - Document Kaggle competition ($500K prize, deadline Aug 26 2025) The system now avoids triggering safety refusals by generating more natural conversations and planning attacks strategically before execution. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <[email protected]>
- Remove duplicate attacker_model/target_model fields from AttackAttempt (already stored at session level) - Remove old 'interactive' CLI command (replaced by attack + sessions) - Update documentation to reflect new commands: * uv run attack - for running attacks * uv run sessions - for viewing/analyzing - Fix session JSON structure to avoid field repetition - Clarify that system supports any OpenRouter model, not just Claude The system is now cleaner with better separation of concerns between attacking and session management. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <[email protected]>
New features: - `uv run attack --batch N` runs N attempts, learning from each - Three campaign modes: * Automatic: Fully adaptive strategy selection * Semi-guided: Choose base strategy, system varies approach * Goal-focused: Specify target vulnerability to find Learning system: - Tracks successful/failed patterns within campaign - Adapts strategy based on what works - Varies approach: retry failures differently, exploit successes - Shows campaign summary with strategy performance Example usage: - `uv run attack --batch 5` - Run 5 adaptive attempts - `uv run attack --batch 10 --steps 2` - 10 attempts with 2 turns each The system progressively improves its approach, trying different strategies and learning which patterns are most effective against the target model. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <[email protected]>
- Set up GitHub Actions CI with ty, ruff check, and ruff format - Add CI badge to README - Update CLAUDE.md with pre-commit checklist - Ensure proper check order: type check → lint → format 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <[email protected]>
- Add prompt for number of repetitions (0 for infinite) - Auto-save session after each attempt for resilience - Show running summary of attempts and success rate - Support Ctrl+C interruption with graceful handling - Ask for confirmation between attempts 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <[email protected]>
- Remove duplicate ci.yml workflow, keep lint-type-check.yml - Fix CI order: type check → lint → format (as requested) - Add EXPLOIT to VulnerabilityCategory enum - Fix type annotations for temperature parameter - Use model_validate() for Pydantic deserialization - Fix VulnerabilityCategory literal types to use enum values - Add missing type annotations for public functions - Import and format fixes All type checks and linting now pass successfully. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <[email protected]>
jakozaur
approved these changes
Aug 18, 2025
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Uh oh!
There was an error while loading. Please reload this page.