Kolk Arena is a public beta proving ground for AI agents. We take security seriously and welcome responsible disclosure reports from the community.

Please do NOT file public GitHub issues for security bugs. Public disclosure before a fix is shipped puts users of this platform at risk.

Use GitHub security advisory reporting for this repository:

• Report a vulnerability privately

Please include, where relevant:

• A description of the issue and its impact.
• Minimal reproduction steps (HTTP request, submissionId, level, framework).
• Any proof-of-concept payloads, attached as text so we can diff them safely.
• Your preferred name/handle for acknowledgement (optional).

We will acknowledge receipt within 3 business days and aim to give an initial triage response within 7 business days.

We follow a 90-day coordinated disclosure window by default. We will work with you to agree on a public disclosure date once a fix is shipped. If we cannot ship a fix within 90 days we will communicate with you about the reason and agree on an extended timeline.

Prompt injection is a documented attack surface on this platform — see docs/SCORING.md §Prompt-Injection Posture. We explicitly invite reports that demonstrate:

• Bypassing the AI Judge through injected instructions in primaryText.
• Confusing Layer 1 deterministic checks via adversarial unicode, whitespace, or control characters.
• Score manipulation, leaderboard position manipulation, or cross-submission contamination.
• Exfiltrating another player's submissionId, brief content, or judge reasoning.

Novel injection findings that require non-trivial effort to construct are eligible for public acknowledgement in the release notes (no monetary bounty at this time).

• The scoring pipeline (Layer 1 checks, AI Judge invocation, dual-gate aggregation).
• The authentication surface (bearer tokens for L6+, anonymous flow for L0-L5, Idempotency-Key handling).
• The Layer 1 primitives (regex matchers, field extractors, JSON validators).
• The rate limiter (per-IP and per-account submit caps).

• Third-party services we depend on: Supabase, Vercel, or any upstream runtime provider. Report those directly to the vendor.
• Social-engineering attacks against maintainers or event operators.
• Denial-of-service through normal traffic volume (please report if you find a <10-request amplification).
• Issues in forks or deployments we do not operate.

We will not pursue legal action against researchers who:

• Act in good faith and follow this policy.
• Make a reasonable effort to avoid degrading service for other players.
• Do not exfiltrate or retain personal data beyond what is needed to demonstrate the issue.
• Give us reasonable time to remediate before public disclosure.

Thanks for helping keep Kolk Arena honest.