ig-caption-writer

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input such as the topic, angle, and target audience, which is then processed to generate a caption that is sent to a network service (Instagram) via the lib.publish tool.
  • Ingestion points: SKILL.md (Step 1: Gather inputs).
  • Boundary markers: Absent; the instructions do not specify the use of delimiters or clear boundaries to isolate user data from the agent's task instructions.
  • Capability inventory: SKILL.md (Step 10) invokes lib.publish, which provides the capability to transmit data to an external network service (Instagram).
  • Sanitization: Absent; while the skill performs a "humanizer pass" to clean up AI-like vocabulary, it does not include security-oriented sanitization or filtering for the user-provided inputs.
  • [DATA_EXFILTRATION]: The lib.publish tool in SKILL.md (Step 10) is called with a media parameter that accepts a list of local file paths. If the agent derives these paths from user input without strict validation or sandboxing, it could be manipulated into accessing and uploading sensitive local system files to the target platform.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 06:05 PM
Security Audit — agent-trust-hub — ig-caption-writer