Back to skill

Security audit

bounty-hunter

Security checks for vulnerabilities and agentic risk

Overview

This toolkit fits its bounty-development purpose, but its broken preflight gate and incomplete test scaffolds require independent verification.

Use this as a development aid, not an automated approval gate. Repair the preflight syntax error, make unavailable guards fail explicitly, replace scaffold placeholders with real assertions, and independently verify failure before the fix and success afterward. The confirmed scanner concerns affect reliability; the artifacts show no unrelated access or hidden execution.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (9)

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 24)May include surrounding context.

md
| `wasm-js-diff-guard.js` | Flags node-supplied hex flowing into a normalizing hash AND a raw `Buffer.from(x,"hex")` with no strict shape check and no shared tr

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 25)May include surrounding context.

md
| `wildcard-ok-guard.js` | Flags state-match `_ =>` arms returning Ok/None (swallowed variants) |

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 26)May include surrounding context.

md
| `mutation-before-fallible-guard.js` | Flags store/cache writes preceding `?`/`await` with no restore on the error path |

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 27)May include surrounding context.

md
| `preflight.mjs` | One-shot gate: `node preflight.mjs payload.json` — files<=3, churn<=80, regression test present, guards clean. Exit 0 = push. |

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 28)May include surrounding context.

md
| `repro-scaffold.mjs` | Harvested `failing_assertion` → drop-in red test (`tests/repro_bounty_<id>.rs` or vitest) |

Diff-hygiene checks block the PR gate despite being documented as warnings.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

L05 explicitly describes file-count and churn limits as 'warn, not fail,' but L21–L23 add them as ordinary checks and L34 requires every check to pass. Consequently, exceeding these limits causes the CLI to exit with failure at L47, contradicting the documented advisory behavior.

Content

No source excerpt is available for this finding.

The documented boundary checks can be silently skipped while the gate reports them as clean.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

Guard imports suppress all loading failures, and preflight skips any unavailable guard while reporting the boundary checks as clean. In a security-oriented PR gate, missing or broken dependencies can therefore produce a successful result without performing the advertised analysis.

Content

No source excerpt is available for this finding.

The TypeScript generator emits an always-passing placeholder despite claiming the red regression test is already written.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The generated TypeScript test succeeds unconditionally and never evaluates the harvested assertion, despite the generator's claim to provide an already-written red test. Although TODO comments disclose that the scaffold is incomplete, accepting this output as regression evidence can let an unfixed security defect pass a test-presence-based workflow.

Content

No source excerpt is available for this finding.

The documented rejecting-catch exemption suppresses findings without checking that the catch rejects.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The shared-try exemption checks only approximate block placement, not whether a catch exists, rejects input, or prevents subsequent use. It then suppresses findings for the entire file, allowing swallowed exceptions, unrelated operations, or non-throwing decoder differences to hide unsafe cross-runtime parsing.

Content

No source excerpt is available for this finding.