connect-me

Warn

Audited by Gen Agent Trust Hub on Sep 23, 2026

Risk Level: MEDIUMPERSISTENCECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PERSISTENCE]: The skill contains detailed instructions and configuration templates in references/keeping-the-replica-running.md for maintaining the catalyst-skills replica process across system restarts. This includes:
  • A launchd plist configuration (dev.catalystcloud.replica.plist) for macOS.
  • A systemd user unit file (catalyst-cloud-replica.service) for Linux.
  • Instructions for creating a Windows Task Scheduler task to run at logon with automatic restart capabilities.
  • [COMMAND_EXECUTION]: The skill uses node:child_process spawn in scripts/lib/cli.mjs and scripts/verify-connection.mjs to execute the catalyst-skills CLI tool. On Windows platforms, the shell: true option is utilized during execution.
  • [EXTERNAL_DOWNLOADS]: The skill invokes npx @catalyst-cloud/catalyst-skills to download and execute the vendor's package from the NPM registry if it is not already installed locally.
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external "tenant contract" data from the Catalyst Cloud service into the agent's context, creating a potential vector for instruction injection if the cloud source is compromised.
  • Ingestion points: Data returned by the catalyst-skills contract command is processed in SKILL.md and scripts/verify-connection.mjs.
  • Boundary markers: Absent from the prompt interpolation logic.
  • Capability inventory: The skill includes file system writes (to ~/.config/catalyst-cloud/) and subprocess execution via spawn.
  • Sanitization: The skill performs JSON parsing but does not include explicit validation or sanitization of the contract's textual content before it is presented to the agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 23, 2026, 08:37 PM
Security Audit — agent-trust-hub — connect-me