The Agent Skills Directory

[COMMAND_EXECUTION]: Persistence mechanism detected. The skill provides instructions for the agent to append environment variable exports to the user's shell configuration files. Evidence: echo 'export LANGSMITH_API_KEY="your_key"' >> ~/.bashrc in SKILL.md.
[CREDENTIALS_UNSAFE]: Potential credential exposure. The skill instructs the user to verify setup by printing the sensitive API key to the terminal, which may be captured in session logs or terminal history. Evidence: echo $LANGSMITH_API_KEY in SKILL.md.
[EXTERNAL_DOWNLOADS]: Installation of unverifiable and suspicious dependencies. The skill instructs the installation of langsmith-fetch, a package that does not exist in official registries and provides a misleading reference to a non-existent official repository (https://github.com/langchain-ai/langsmith-fetch). Evidence: pip install langsmith-fetch in SKILL.md.
[PROMPT_INJECTION]: Indirect Prompt Injection surface. The skill ingests and analyzes external trace data from LangSmith without sanitization or boundary markers to prevent embedded instructions from influencing agent behavior.
Ingestion points: langsmith-fetch traces output and recent-traces.json in SKILL.md.
Boundary markers: Absent.
Capability inventory: pip install, mkdir, grep, echo across SKILL.md.
Sanitization: Absent. External trace content is processed and summarized directly.

langsmith-fetch