The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow section in SKILL.md instructs the agent to "Run the smallest safe example or relevant check when feasible." This directs the agent to execute commands or scripts present in the repository being processed.- [REMOTE_CODE_EXECUTION]: By instructing the agent to execute code from the project environment, the skill creates a path for executing potentially malicious code found in external or untrusted repositories.- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted sources (repository files, manifests, and tests) without explicit sanitization or boundary markers, creating a surface where malicious content could influence agent actions, particularly during the verification step.
Ingestion points: Repository files (README, manifests, APIs, examples, tests) as described in the Workflow section of SKILL.md.
Boundary markers: Absent.
Capability inventory: The agent is authorized by the instructions to run examples and checks.
Sanitization: Absent.

create-readme