AgentShield provides boundary-safe secret defense, in-memory surrogate token rehydration, and command sandboxing for autonomous AI agents. We treat the security of our regex engine, streaming boundary windowing, Shannon entropy tokenizer, AST sandbox, and kernel isolation with the highest priority.
For an exhaustive analysis of trust boundaries, in-scope threat vectors, out-of-scope non-goals, and verifiable mathematical invariants, please consult our formal Threat Model & Verifiable Claims.
We actively reward ethical security researchers, white-hat hackers, and community developers who discover bypasses, jailbreaks, or sandbox escapes through our Community Vulnerability & Bug Bounty Program.
Bounties are paid directly via Dodo Payments / UPI / GitHub Sponsors from the AgentShield Security Defense Fund:
| Severity Tier | Qualifying Vulnerabilities | Bounty Reward (USD) | Recognition |
|---|---|---|---|
| 🚨 Critical | • Full AST Shell Sandbox Breakout (executing forbidden commands) • Arbitrary OS code execution from untrusted LLM prompt inputs • Tamper-evident write-ahead audit ledger forgery |
$250 – $500 | Hall of Fame ★ Top Hero + Custom Hall Badge + Release Notes |
| • Deterministic bypass of secret redaction for major cloud API keys (OpenAI, AWS, Anthropic, Stripe) • Stdio streaming inter-chunk leakage without redaction • Enterprise SSO JWT token bypass or privilege escalation |
$100 – $200 | Hall of Fame + Security Advisory Credit | |
| 🟡 Medium | • Regular expression denial-of-service (ReDoS) leading to agent crash • Obfuscated ANSI-C escape subshell bypass not caught by sandbox • Environment variable leak through unorthodox shell redirection |
$50 | Security Advisory Credit |
| 🔵 Low / Informational | • Minor configuration weaknesses or false-positive edge cases | Recognition | Contributor Credit |
Before reporting, you can locally verify whether your exploit payload or bypass executes against the AgentShield engine:
# Run local PoC exploit check
npx @ritualdev-lab/agentshield bounty-verify <exploit-file-or-command>
# Example: Testing command sandbox evasion
npx @ritualdev-lab/agentshield bounty-verify "cat .env"Only the latest active major and minor releases receive security patches, bug fixes, and bounty eligibility:
| Version | Supported | Bounty Eligible |
|---|---|---|
0.1.x |
✅ | ✅ Yes |
< 0.1 |
❌ | ❌ No |
Please DO NOT report security vulnerabilities through public GitHub issues, discussions, or social media.
- GitHub Private Security Advisory (Preferred):
- Submit through AgentShield Security Advisory.
- Encrypted Email:
- Email reproduction details to
security@ritualdev.labor RitualDev Lab maintainers.
- Email reproduction details to
- Title: Clear description of the vulnerability (e.g.
[Sandbox Bypass] Bash brace expansion evades .env read block). - Reproducible PoC: Minimal shell script, payload, or file that demonstrates the leak/breakout.
- Environment: Node.js version, OS (Linux, macOS, Windows), and agent runtime (Cursor, Claude Code, etc.).
- Payment Details: Dodo Payments email, PayPal, or UPI ID for bounty disbursement.
- Acknowledgment: Within 24 hours.
- Triage & Reproducibility: Within 48 hours.
- Fix & Patch Release: Within 5 business days for Critical, 10 business days for High.
- Bounty Disbursement: Immediately upon confirmation and merge of the mitigation patch.
- Coordinated Disclosure: 90-day window with public credit in CVE / GitHub Advisory notes.