The Agent Skills Directory

[PERSISTENCE]: The skill provides instructions to modify the user's shell configuration files (~/.bashrc and ~/.zshrc) to permanently store environment variables using the echo ... >> command. This modification ensures that specific configurations and credentials persist across all future terminal sessions.
[CREDENTIALS_UNSAFE]: The documentation includes steps to verify the environment by printing the LANGSMITH_API_KEY to the terminal. If performed by an automated agent, this action exposes sensitive credentials within the conversation history, terminal logs, or system output.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze execution traces and thread data from an external platform (LangSmith). This represents a vulnerability surface where untrusted data from previous executions could contain instructions that attempt to influence the agent's behavior during the analysis process.
Ingestion points: Trace data fetched via langsmith-fetch traces and conversation threads via langsmith-fetch threads.
Capability inventory: Shell command execution (CLI tools) and file system operations (creating session directories and writing trace files).
Boundary markers: No specific delimiters or instructions are used to isolate or ignore instructions embedded within the fetched external data.
Sanitization: The skill relies on basic pattern matching (grep) without explicit validation or escaping of the retrieved content.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the langsmith-fetch Python package from a repository associated with a known organization. This package provides the core functionality needed for the skill's debugging workflows.

langsmith-fetch