The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions in shared/evals/build-eval.md and shared/evals/eval-hillclimb.md that advise the AI agent to be concise and avoid narrating internal bookkeeping steps unless explicitly asked by the user. While this matches some concealment detection heuristics, it is a standard design pattern for improving user experience in complex agentic workflows and does not appear to be used for malicious purposes.
[EXTERNAL_DOWNLOADS]: The documentation contains legitimate installation commands for the Anthropic CLI and various official SDKs. It also references httpx2, which is described as a maintained fork of the httpx library used by the Anthropic Python SDK. All sources cited are official project repositories or standard package registries.
[COMMAND_EXECUTION]: The reference material includes numerous examples of shell commands for the ant CLI and curl. These are provided for educational and developer utility purposes, enabling users to test API interactions and manage resources like Agents and Environments.

claude-api