The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by interpolating untrusted user data into its structured output artifacts.
Ingestion points: The variables user_request and context.constraints from the input object in src/index.js are used to construct the final output artifacts.
Boundary markers: Absent; the skill does not use XML tags, dedicated block delimiters, or specific instructions to isolate user-provided text from the analysis framework.
Capability inventory: The skill possesses no network, file-system, or dynamic execution capabilities within its primary logic. A separate validation script uses subprocess execution on generated output.
Sanitization: There is no evidence of input validation or character escaping for user-provided strings before they are rendered in the output brief.
[COMMAND_EXECUTION]: The skill includes a helper script, scripts/validate_reasoning_artifacts.py, which executes a local Python utility via subprocess.run to validate the structure of reasoning artifacts.
The execution targets an internal validator located at a relative path within the vendor's environment.
Arguments passed to the subprocess are restricted to static validation parameters and the path to the generated output.

deconstruction-master