privacy-publish

Pass

Audited by Gen Agent Trust Hub on Aug 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes shell commands using the Bash tool to run a local Python script (asc.py) and various external hosting CLIs (git, Netlify, S3). This provides the agent with the ability to execute code and interact with the file system and network.- [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted data from local markdown files (.planning/legal/privacy.md, .planning/legal/terms.md) and potentially external web pages via claude-in-chrome.
  • Ingestion points: Markdown files in .planning/legal/ and external pages visited during the publishing flow.
  • Boundary markers: None identified; the agent is not explicitly instructed to ignore instructions embedded within the legal drafts.
  • Capability inventory: The agent can execute shell commands and perform browser automation.
  • Sanitization: No evidence of input sanitization or validation before processing the documents or using the data to construct API calls.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 2, 2026, 01:19 PM
Security Audit — agent-trust-hub — privacy-publish