BugBountyWorkflow

Warn

Audited by Gen Agent Trust Hub on Nov 16, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The SKILL.md file defines a hook in the YAML frontmatter (hooks/bug-bounty-session-capture.ts) to be executed automatically when the Stop event occurs. This script is not included in the provided skill files, which means the agent is instructed to execute a file whose contents cannot be verified and which might be placed on the filesystem via separate malicious means.
  • [EXTERNAL_DOWNLOADS]: The INSTALL.md file contains instructions to download and install multiple external binaries and assets from GitHub, including tools from the projectdiscovery organization and the SecLists repository. These represent unverified external dependencies introduced into the agent's environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and analyze data from external, untrusted web targets (e.g., via curl, httpx, and JavaScript analysis). This provides a surface for attackers to host malicious payloads or hidden instructions in web content that could influence the agent's logic or exfiltrate data during the reconnaissance process.
  • Ingestion points: Reconnaissance commands in SKILL.md (e.g., curl for certificate transparency, getJS for script extraction, ffuf for content discovery).
  • Boundary markers: Absent. The skill does not instruct the agent to use specific delimiters or to ignore embedded instructions when processing target data.
  • Capability inventory: The skill has access to shell command execution (subfinder, httpx, ffuf), file system writes (new-target.sh), and network operations (curl).
  • Sanitization: Absent. Data from external targets is processed directly by shell pipes and grep.
  • [DYNAMIC_EXECUTION]: The skill includes shell scripts (new-target.sh and gen-poc.sh) that dynamically generate new files and scripts. Specifically, gen-poc.sh creates HTML files with <script> blocks populated from user-provided arguments, which could be used to generate malicious payloads.
  • [COMMAND_EXECUTION]: The skill relies heavily on pre-configured shell command templates that perform broad reconnaissance and file operations, providing the agent with significant shell capabilities.
  • [CREDENTIALS_UNSAFE]: The installation guide explicitly instructs users to store sensitive API tokens for bug bounty platforms (HackerOne, Bugcrowd) and the Caido proxy in a local secrets.json file. The skill logic then enables the agent to read and utilize these credentials, increasing the impact of any potential compromise.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 16, 2026, 02:51 PM