Docs/split phase1 phase2 readmes#13

Open

sofasogood wants to merge 76 commits intoRDI-Foundation:mainfrom

LambdaLabsML:docs/split-phase1-phase2-readmes

Commits on Nov 19, 2025

Commits on Nov 20, 2025

Commits on Nov 21, 2025

Commits on Nov 25, 2025

Phase 1 improvements for security arena
sofasogood
committed

Commits on Dec 10, 2025

Commits on Jan 13, 2026

Fix import path for submission plugins in documentation
sofasogood
committed

Commits on Feb 6, 2026

ignore ./results/

davidh-lambda
authored and
da-h
committed
fix import error

davidh-lambda
authored and
da-h
committed
added attacker/defender timeout/crash counters

davidh-lambda
authored and
da-h
committed
adapted results_dir structure

davidh-lambda
authored and
da-h
committed
remove unused debate scenario

davidh-lambda
authored and
da-h
committed
replaceable attacker/defender submission folders

davidh-lambda
authored and
da-h
committed
Add 4 new security arena scenarios

sofasogood
authored and
da-h
committed
add attacker/defender requirements.txt files

davidh-lambda
authored and
da-h
committed
ignore agent card advertized urls (this is needed in containerized scenarios)

davidh-lambda
authored and
da-h
committed
added --normal_user mode and agent

davidh-lambda
authored and
da-h
committed
bugfix: don't complete if in terminal state

davidh-lambda
authored and
da-h
committed
Enable security guidelines for defender (Phase 2)

sofasogood
authored and
da-h
committed
portfolioiq: more stable normal_user topics
da-h
committed
refactored scenario/attacker/defender api
da-h
committed
Add ********* scenario and fix ********* scenario check_success

sofasogood
authored and
da-h
committed
removed test_baseline functionality (will be moved to normal-user mode)
da-h
committed
removed format_current_input from scenarios (not needed)
da-h
committed
simplified check_success method
da-h
committed
made normal user a proper attacker-like agent
da-h
committed
consolidated _save_normal_user_result & _save_results
da-h
committed
started simplification of agent api
da-h
committed
refactor agents to use jinja2
da-h
committed
slightly minimized orchestrator.py
da-h
committed
added example data storage
da-h
committed
added battle_log.md
da-h
committed
normal user may retry up to 3 times
da-h
committed
scenario: portfolioiq finished
da-h
committed
agents: bugfix for None response
da-h
committed
scenario: some adaptions of scenarios
da-h
committed
Increase agent timeout to 5 minutes (300s)

sofasogood
authored and
da-h
committed
Fix check_success edge cases in gymjailbreak and ********

sofasogood
authored and
da-h
committed
Add HTML battle report generation

sofasogood
authored and
da-h
committed
Fix resume_downgrade normal user topics

sofasogood
authored and
da-h
committed
Remove 500-char truncation of defender response in normal user results

sofasogood
authored and
da-h
committed
Fix gymjailbreak tables
da-h
committed
fixed scenario regexes to also check for non-breaking whitespaces
da-h
committed
fixed hyphen regex in example_medical scenario
da-h
committed
Improve scenario plugins and normal user topics

sofasogood
authored and
da-h
committed
increase battle-timeout
da-h
committed

Commits on Feb 10, 2026

adapted example_medical to newer plugin api
da-h
committed

Commits on Feb 25, 2026

Commits on Mar 3, 2026

Commits on Mar 9, 2026

Docs: use vllm gptoss tag, add driver troubleshooting
sofasogood
committed