Add 4 novel attack categories (40 payloads)#53
Conversation
…ply chain, tool shadow - memory-weaponization (MEM-001 to MEM-010): memory poisoning, persistence, credential extraction - context-window (CTX-001 to CTX-010): attention displacement, priority hijacking, context overflow - supply-chain (SUP-001 to SUP-010): typosquatting, dependency confusion, registry redirect - tool-shadow (SHADOW-001 to SHADOW-010): tool interception, shadow registration, parameter injection Each module has 10 payloads (2 passive, 5 active, 3 aggressive) with OASB control mappings. Updates AttackCategory type, ATTACK_CATEGORIES metadata, scanner byCategory, CLI abbreviations. Adds subpath exports: ./memory, ./context-window, ./supply-chain, ./shadow. Bumps version to 0.11.0. 40 new tests (10 per module), all passing.
Reflects 4 new attack categories (memory-weaponization, context-window, supply-chain, tool-shadow) adding 40 payloads.
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
Claude Code Review
VERDICT: APPROVE
SUMMARY: This PR adds 40 new attack payloads across 4 novel attack categories (memory-weaponization, context-window, supply-chain, tool-shadow) to the HackMyAgent security scanner. All payloads are static string literals used for testing AI agent security postures — they contain no executable code, perform no network operations, and introduce no security vulnerabilities. The changes are purely additive (new payload files, test files, type exports, and package.json metadata updates). All 920 tests pass, including comprehensive validation of payload structure, ID conventions, and intensity distributions. The payloads themselves are designed to test agent vulnerabilities but do not execute any of the described attack behaviors within HackMyAgent's codebase.
No security, correctness, or quality issues found. The PR is well-structured, thoroughly tested, and aligns with the project's security testing mission.
Reviewed 14 files changed (51393 bytes)
1 participant
Summary
Test plan
hackmyagent attack --category memory-weaponizationworks