social-har-api-connectivity

Pass

Audited by Gen Agent Trust Hub on Sep 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/chrome_capture_client.py utilizes the subprocess.Popen function to launch the Google Chrome browser. By default, it targets a hardcoded path on macOS (/Applications/Google Chrome.app/Contents/MacOS/Google Chrome) and applies various command-line flags to enable remote debugging via CDP and manage a temporary user data directory.
  • [EXTERNAL_DOWNLOADS]: The skill requires the installation of the websockets Python package to facilitate communication with the Chrome DevTools Protocol. This is a standard dependency for the skill's network capture functionality.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process raw network traffic (HAR files) from external websites. This content is subsequently summarized and presented to the agent. Because the traffic originates from untrusted external sources, malicious content embedded in webpage headers or response bodies could potentially serve as a prompt injection intended to influence the agent's reasoning or subsequent actions.
  • Ingestion points: Network event data captured by scripts/chrome_capture_client.py and HAR files processed by scripts/har_derive.py.
  • Boundary markers: None identified; the agent reads tool output containing raw strings from the capture.
  • Capability inventory: File system writing (os.chmod), subprocess execution (launching Chrome), and network monitoring.
  • Sanitization: scripts/har_derive.py implements a safe_excerpt function that uses regular expressions to redact sensitive tokens (passwords, secrets) from output, though it does not filter for prompt injection patterns.
  • [DATA_EXFILTRATION]: The primary purpose of this skill is the automated extraction of sensitive authentication data, including session cookies and authorization headers. The skill incorporates several safety measures: it saves these credentials to a local temporary directory with restricted permissions (chmod 600) and includes explicit instructions for the user to avoid committing these files to version control repositories.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 2, 2026, 09:39 PM
Security Audit — agent-trust-hub — social-har-api-connectivity