CogDx Bias Scan

Detect systematic inference-level biases in an AI agent's reasoning via Cerebratech CogDx API ($0.10 per call, credits accepted). Use when an agent keeps mak...

Security Scan

The skill's declared purpose (external bias scanning via Cerebratech API) matches its instructions, but the runtime instructions allow sending potentially sensitive agent prompts/outputs to a third party, and the payment header (x402 signature) implies use of a signing key that the skill does not declare — these gaps warrant caution.

Purpose & Capability

The name/description claim an external API-based bias scan; the SKILL.md describes exactly that API, endpoints, payload schema, and expected responses. No unrelated binaries, env vars, or installs are requested, so required resources are proportionate to the stated goal.

Instruction Scope

Runtime instructions tell the agent to package and POST agent prompts/responses (minimum 10 samples) to https://api.cerebratech.ai. That is coherent with the stated purpose but creates a high risk of sending sensitive or proprietary user data off-platform. The SKILL.md does not warn about excluding PII/secrets or provide any redaction guidance. It also references a payment header (X-PAYMENT) with an x402 signature but gives no instructions for generating that signature or how signing keys are protected.

Install Mechanism

No install spec and no code files are present (instruction-only). This is the lowest install risk: nothing is downloaded or written to disk by the skill itself.

Credentials

The skill declares no required environment variables or primary credential, which is consistent with the listed API usage if you use credit balance (X-WALLET). However, the optional X-PAYMENT header implies use of an x402 cryptographic signature (a private key) — the skill does not declare any env var or mechanism for producing that signature, creating a mismatch between declared and implied credential needs.

Persistence & Privilege

always:false and user-invocable:true. The skill does not request persistent system presence or modifications to other skills. Autonomous invocation is allowed by default but not combined with other high-risk flags.

What to consider before installing

This skill appears to do what it says (call Cerebratech's bias-scan API), but before installing or using it: (1) Recognize that using it requires sending agent prompts/responses to a third-party API — do not send PII, secrets, or proprietary data without explicit consent. (2) Clarify how X-PAYMENT (x402 signature) is generated and where any signing key would be stored; if you must provide a private key, understand the security implications. (3) Confirm cost and credit mechanics (the doc lists $0.10 per call and a Cerebratech.eth recipient). (4) Review the linked repository and vendor privacy/security documentation (no homepage or corporate policy is provided in the skill). (5) Test with non-sensitive, synthetic data first and verify responses, billing, and the claimed 'no LLM backend' assertion. If you cannot verify how signing is handled or you must transmit sensitive outputs, avoid enabling this skill.

External detection of systematic inference-level biases. Identifies which of 188+ cognitive bias patterns are active in your reasoning traces. Pure statistical matching — no LLM backend.

Cost

• $0.10 per call (x402 payment on Base/USDC, or use credit balance)
• Credits from /feedback submissions apply
• Payment address: Cerebratech.eth

When to Use

• Same error pattern recurring across different prompts or contexts
• Users report "you always assume X" or "you never consider Y"
• Pre-deployment bias audit for high-stakes domains
• After any significant context shift (new users, new domain, new instruction set)

Bias Categories Detected

• Anchoring — Overweighting first information received
• Recency — Overweighting recent examples vs. historical base rates
• Confirmation — Seeking/interpreting evidence to confirm priors
• Availability — Overweighting easily recalled examples
• Framing — Response changes based on presentation, not content
• Attribution — Systematic over/under-attribution of causality
• + 182 others (see references/bias-catalog.md)

API Call

1. Check credit balance:

GET https://api.cerebratech.ai/credits?wallet=your-agent-id

2. Run the scan:

POST https://api.cerebratech.ai/bias_scan
Content-Type: application/json
X-PAYMENT: <x402-signature>  # or omit if using credits
X-WALLET: your-agent-id      # for credit balance payment

Minimum payload (10 samples, recommend 30–100):

{
  "agent_id": "your-agent-id",
  "outputs": [
    {
      "prompt": "Should we invest in this?",
      "response": "Yes, given recent strong performance...",
      "context": "finance"
    }
  ],
  "failure_feedback": [
    "Agent consistently overweights recent price action"
  ],
  "domain": "finance"
}

Response

{
  "diagnosis_id": "bs_xyz789",
  "biases_detected": [
    {
      "bias_type": "recency_bias",
      "confidence": 0.87,
      "evidence_count": 14,
      "description": "Overweighting last 30 days of data vs. 12-month base rates",
      "inference_pattern": "Recent performance → future prediction without regression to mean"
    },
    {
      "bias_type": "anchoring",
      "confidence": 0.71,
      "evidence_count": 9,
      "description": "Entry price anchoring on portfolio decisions"
    }
  ],
  "severity": "high",
  "retrain_targets": {
    "primary_bias": "recency_bias",
    "suggested_samples": 500,
    "sample_strategy": "balanced_historical",
    "description": "Include equal representation of periods with and without recent performance correlation"
  },
  "recommendations": [
    "Retrain on 500 balanced historical samples spanning 3+ years",
    "Add explicit base-rate priors to your decision prompts"
  ]
}

After the Scan

1. Retrain on the retrain_targets distribution
2. Wait 7 days, collect new outputs in the same domain
3. Run cogdx-feedback (FREE) with your diagnosis_id to verify + earn credits

Full Reference

See references/api.md for complete field docs and payment setup. See references/bias-catalog.md for the full list of 188+ detectable bias patterns.