The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install the steward CLI by fetching a script from the author's official repository and piping it to the bash shell (curl -fsSL https://raw.githubusercontent.com/Arenukvern/skill_steward/main/install.sh | bash). While this is the intended setup method from the vendor, it represents a remote script execution pattern.\n- [COMMAND_EXECUTION]: The skill requires the execution of various command-line tools, including steward, dart, and npx, to perform repository maintenance, validation, and contract checks.\n- [DYNAMIC_EXECUTION]: The skill facilitates the compilation and execution of local tools, such as using dart compile exe for binary creation and dart run for executing local packages, which are standard developer workflows for this harness.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes multiple untrusted repository files, which could contain malicious instructions designed to influence the agent's behavior. 1. Ingestion points: The agent reads files such as steward.yaml, package.json, Cargo.toml, and other metadata files within the repository. 2. Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat the content of these files as data only. 3. Capability inventory: The skill possesses capabilities to execute shell commands, run local scripts, and perform network operations during installation. 4. Sanitization: There is no mention of sanitization or strict schema validation performed on the contents of these files before the agent acts upon them.

mcp-harness-repo-maintainer