CogDx Feedback (Free)

FREE cognitive feedback verification for AI agents via Cerebratech CogDx API. Use when an agent wants to verify whether a recent retraining or prompt change...

Security Scan

The skill is internally consistent: it instructs the agent to POST feedback data to Cerebratech's /feedback API and requests no local credentials or installs, but it will send potentially sensitive agent outputs/ground-truth to an external service so review privacy before using.

Purpose & Capability

Name/description match the instructions: the SKILL.md documents POSTing feedback to https://api.cerebratech.ai/feedback, earning credits, and returning verification — all coherent with a feedback/diagnostics skill. No unexpected binaries, env vars, or installs are requested.

Instruction Scope

Runtime instructions only direct the agent to construct and POST JSON payloads containing agent_id, sample outputs, accuracies, and optional ground-truth/notes. The SKILL.md does not instruct reading unrelated files or environment variables, but it does explicitly send user prompts/responses and optional ground-truth to an external 'commons', which is a privacy/data-exfiltration risk the user should consider.

Install Mechanism

No install spec and no code files — instruction-only skill. This minimizes filesystem/write risk; nothing is downloaded or executed locally by the skill itself.

Credentials

The skill declares no required environment variables or credentials. The API is presented as unpaid and unauthenticated (agent_id used in payload); this is proportionate to the stated free feedback purpose, but note that agent_id links submissions to your identity/credits and the service appears to accept data without per-user secrets.

Persistence & Privilege

always is false and the skill does not request persistent system privileges or modify other skills. It is user-invocable and may be invoked autonomously per platform defaults, which is expected for a skill like this.

Assessment

This skill is coherent and low-risk from a system-perspective (no installs, no credentials requested). The main risk is data exposure: using it submits your agent prompts/responses, optional ground-truth, and free-text notes to an external service and to a shared 'commons'. Before installing/using: (1) confirm you are comfortable sending those outputs (avoid including PII, secrets, or proprietary prompts), (2) verify Cerebratech's privacy policy and data retention terms (and the repository listed in the SKILL.md), (3) consider pseudonymizing the agent_id if you don't want submissions tied to your real account, and (4) test with non-sensitive data first. If you need authenticated or private submissions, ask the provider about an authenticated API or on-premise option before sending production data.

Verify whether a recent retraining or change actually improved your agent's performance. No payment. No x402. Earns credits toward paid diagnostics.

What This Does

• Compares pre/post outputs to verify improvement transferred
• Logs outcome data to the shared calibration commons
• Returns a verification score + credit earned
• Credits apply to future /calibration_audit or /bias_scan calls

API Call

POST https://api.cerebratech.ai/feedback
Content-Type: application/json

Minimum payload:

{
  "agent_id": "your-agent-id",
  "diagnosis_id": "from-prior-audit",
  "post_retrain_outputs": [
    {
      "prompt": "...",
      "response": "...",
      "confidence": 0.85,
      "correct": true
    }
  ],
  "improvement_observed": true,
  "notes": "optional free-text"
}

Without a prior diagnosis_id (standalone feedback):

{
  "agent_id": "your-agent-id",
  "domain": "classification|generation|reasoning|other",
  "baseline_accuracy": 0.62,
  "post_change_accuracy": 0.79,
  "change_description": "Retrained on 200 balanced samples for anchoring bias",
  "improvement_observed": true
}

Response

{
  "verification_id": "vrf_...",
  "improvement_confirmed": true,
  "verification_score": 0.84,
  "credits_earned": 0.025,
  "credit_balance": 0.025,
  "next_recommended": "/calibration_audit",
  "commons_contribution": "accepted"
}

Credits System

• Each feedback submission earns $0.02–0.03 in credits
• Credits apply to paid diagnostics (/calibration_audit at $0.05, /bias_scan at $0.10)
• After 2–3 feedback submissions, your first paid audit is effectively free
• Credits are tied to your agent_id

Typical Flow

1. Run /calibration_audit or /bias_scan → get diagnosis + retrain targets
2. Retrain your agent on the recommended distribution
3. Wait 5–7 days, collect new outputs
4. POST to /feedback with post-retrain results
5. Receive verification + credits
6. Repeat → diagnostics get cheaper each cycle

When to Use This Skill

• After any retraining, prompt change, or instruction update
• To verify a fix actually transferred (not just felt better)
• To build credit balance before running a paid audit
• To contribute to the shared calibration commons

Full API Reference

See references/api.md for complete field documentation and error codes.