The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted external data from the project codebase, including log files (.log), data exports (.csv, .jsonl), and documentation (CLAUDE.md).
Ingestion points: Project files including .json, .csv, .jsonl, .log, and .env are searched and parsed to identify tracing setups and log volume.
Boundary markers: None identified. The skill does not implement specific delimiters or 'ignore' instructions when reading content from these files.
Capability inventory: The skill has the capability to read local files and search the directory structure. It does not possess network-write or external command execution capabilities.
Sanitization: None identified. The content of the logs and project files is used directly to determine readiness status.
[DATA_EXFILTRATION]: The skill instructs the agent to search for sensitive files and environment variables, specifically targeting .env files and the LATITUDE_API_KEY.
Evidence: The instructions explicitly ask the agent to 'Search the codebase' for 'Environment variables referencing observability: LATITUDE_API_KEY... or any logging... to a file, database, or external service.'
Context: As the skill is authored by the vendor associated with these keys (latitude-dev), this behavior is consistent with a readiness check for their own platform. There is no evidence of network instructions to exfiltrate these keys to an unauthorized third party.
[SAFE]: The skill references external documentation and tools from well-known and trusted services.
Evidence: Links to docs.latitude.so and references to tools like opentelemetry, langsmith, and braintrust are used appropriately for their intended purpose in observability workflows.

llm-evals-checklist