gh-issue-to-demand-signal

Warn

Audited by Gen Agent Trust Hub on May 17, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The README.md file contains installation instructions using npx "@opendirectory.dev/skills". This command downloads and executes code from a third-party NPM scope that is not identified as a trusted organization or well-known service.
  • [COMMAND_EXECUTION]: The skill uses dynamic script generation via Python HEREDOCs (e.g., in Step 2 of SKILL.md). It interpolates user-provided repository URLs directly into a Python string literal: raw = "REPO_INPUT_HERE". If the agent does not properly sanitize the user input, an attacker could provide a string that escapes the literal context (e.g., using quotes and semicolons) to execute arbitrary Python code.
  • [PROMPT_INJECTION]: The skill exhibits a significant surface for indirect prompt injection by ingesting and processing external data.
  • Ingestion points: GitHub REST API (Step 3) fetches issue titles and bodies which are stored in /tmp/ghd-raw-issues.json and subsequently processed in Steps 4, 5, and 6.
  • Capability inventory: The skill executes shell commands (bash), runs Python scripts, and writes output to the local file system (docs/demand-signals/).
  • Boundary markers: There are no explicit delimiters or instructions provided to the AI to ignore instructions embedded within the fetched GitHub issue content.
  • Sanitization: No evidence of sanitization or escaping of the issue content is present before it is interpolated into prompts for clustering and GTM messaging generation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 17, 2026, 01:46 PM
Security Audit — agent-trust-hub — gh-issue-to-demand-signal