trail
Warn
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: MEDIUMINDIRECT_PROMPT_INJECTIONPERSISTENCECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions include a process for downloading and installing the '@trailhq/trail-code' package via npm when the tool is missing. This is a vendor-controlled package.
- [COMMAND_EXECUTION]: The skill uses multiple shell commands, including
npm,git, and various subcommands of thetrailCLI such asask,grep,skeleton,callers,map,build,check, andnote. - [PERSISTENCE]: The skill directs the agent to offer modifications to the user's shell profile to update the PATH environment variable with
~/.npm-global/bin. This ensures the installed tools are available in future sessions. - [INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes 'learnings' from markdown files in the
.trail/learnings/directory to provide context for tasks. - Ingestion points: Markdown files in the
.trail/learnings/directory. - Boundary markers: The skill includes instructions to treat these files as historical context and not as direct commands to be followed.
- Capability inventory: The agent can install packages, modify shell profiles, and perform git operations.
- Sanitization: No explicit content validation or sanitization for the learning files is defined.
Audit Metadata