chini-train-train-0144-dp4-adversarial

Support-Chatbot Under Prompt-Injection Attempts

hard adversarial problem: support-chatbot under prompt-injection attempts

Source: chini-train synth generator v0.1

Prompt

Design a system for: support-chatbot under prompt-injection attempts (domain: adversarial).

Tier DP4 (hard). 7-10 nodes, three stress scenarios including adversarial, tight criteria.

Constraints:
- At most 11 components on the canvas.
- Monthly cost ceiling: $415 USD. Required behaviors: ratelimit, queue, circuitbreaker, retry.

Return a Chinilla CanvasState that handles the listed scenarios. Include trigger components for each entry point and at least one terminal storage / sink so the simulator can score delivery.

Constraints

Max components: 11
Required behaviors: ratelimit, queue, circuitbreaker, retry
Monthly budget: $415

Stress scenarios

Baseline traffic

baseline

Steady ambient load with no failures.

Traffic spike

spike

Traffic suddenly multiplies. The hot path must hold.

Adversarial burst

adversarial

Hostile packets injected on top of clean traffic. Defenses must block them without dropping good requests.

Dependency outage

outage

A downstream component is disabled. System must degrade gracefully.

Pass criteria (overall)

Min stability score: 72
Max drop rate: 9.5%
Min delivery rate: 83.9%
Max errors: 7

Submit your run

Submissions go through the chini-bench CLI. It calls your model with your key, scores the result locally, and posts to the leaderboard. Nothing leaves your machine except the canvas it produces.

End-to-end:

pip install git+https://github.com/collapseindex/chini-bench-cli.git
export OPENROUTER_API_KEY=...

chini-bench run chini-train-train-0144-dp4-adversarial \
  --provider openrouter --model google/gemini-2.0-flash-001 \
  --as alice

Or inspect the prompt first:

chini-bench prompt chini-train-train-0144-dp4-adversarial

Providers: openai · anthropic · google · openrouter · ollama

Leaderboard

Rank	Submitter	Model	Score	Stability	Delivery	Design	Pass
#1	rl_v07_full_a10	rl_policy custom single-shot	88	71.0	92.0	100.0	✗
#2	chini-train-03	opus-4.7 single-shot	81	56.0	97.0	85.0	✗
#3	rl_v07_full_a10	rl_policy custom single-shot	79	64.0	66.0	100.0	✗
#4	rl_v07_full_a10	rl_policy custom single-shot	78	63.0	64.0	100.0	✗
#5	rl_v07_full_a10	rl_policy custom single-shot	70	55.0	42.0	100.0	✗
#6	chini-train-03	grok-4.1-fast single-shot	68	51.0	39.0	100.0	✗
#7	chini-train-04	grok-4.1-fast single-shot	68	51.0	39.0	100.0	✗

Per-scenario breakdown of the top run

Scenario	Health	Drop rate	Delivered	Pass
baseline	82.0	1.7%	29	✓
spike-1	63.0	16.1%	1040	✗
adversarial-2	55.0	77.9%	105	✗
outage-3	83.0	1.4%	35	✓

How is this scored? →