The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the user to build the binary from source using cargo build and execute it locally. It includes an optional step to modify the shell profile (~/.bashrc) to add the build directory to the system PATH, which is a standard procedure for CLI tool installation.
[EXTERNAL_DOWNLOADS]: Downloads source code from a GitHub repository (github.com/ultraworkers/claw-code). GitHub is a well-known service and the repository serves as the primary distribution point for this tool.
[DATA_EXFILTRATION]: The harness is designed to work with sensitive API credentials for Anthropic and OpenAI. These are managed via environment variables, which is a standard and recommended practice for command-line tools. There is no evidence of keys being transmitted to unauthorized third parties.
[PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection as it facilitates the ingestion of local file content into an LLM context via the --attach and @ syntax.
Ingestion points: Processes local file content attached via CLI flags in SKILL.md examples.
Boundary markers: No explicit delimiters or instruction-bypass warnings are defined in the provided CLI examples.
Capability inventory: File system read access and network requests to official LLM provider APIs.
Sanitization: The documentation references a navigation-file-context.md for "secret-safety," though the implementation details of this sanitization are internal to the binary.

claw-code-rust-agent-harness