pp-instacart

Warn

Audited by Gen Agent Trust Hub on May 8, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The "Backfill Flow" instructs the agent to download three JavaScript files (dumper.js, extract-one.js, and export-jsonl.js) from a remote GitHub repository and execute them in the context of the user's browser session using the mcp__claude-in-chrome__javascript_tool. This technique executes remote code within a sensitive, authenticated web environment.
  • [CREDENTIALS_UNSAFE]: The CLI tool is designed to harvest session cookies directly from the user's Chrome browser profile via the instacart auth login command. These credentials are saved to a local configuration file at ~/.config/instacart/session.json, creating a risk of credential exposure if the local environment is compromised.
  • [EXTERNAL_DOWNLOADS]: The skill requires installing a third-party binary tool. It provides instructions to download and install instacart-pp-cli via npm (@mvanhorn/printing-press) or by using go install from the author's GitHub repository.
  • [COMMAND_EXECUTION]: The skill heavily relies on the Bash tool to run the instacart-pp-cli binary, manage software installation, and perform file system operations such as importing order history from JSONL files.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing and caching external data from Instacart order history.
  • Ingestion points: instacart-orders.jsonl (SKILL.md)
  • Boundary markers: Absent.
  • Capability inventory: Subprocess execution via Bash tool for cart modifications and product searches.
  • Sanitization: No explicit sanitization or validation of the ingested JSONL content is described in the provided instructions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 8, 2026, 06:02 PM