The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interpolates the target spec input directly into a shell command in SKILL.md: npx jest --testPathPattern="engine\\.<spec>". This creates a command injection vulnerability where a user can provide crafted input containing shell metacharacters (e.g., ;, &&, or |) to execute arbitrary commands on the host system.\n- [REMOTE_CODE_EXECUTION]: The skill facilitates arbitrary code execution by taking a user-provided Zod snippet and writing it into a .spec.ts file, which is then executed by the test runner. A crafted snippet can escape the expected Zod construction to execute arbitrary system commands during the verification step (Step 4).\n- [PROMPT_INJECTION]: The skill exhibits an unsafe data processing surface for indirect injection.\n
Ingestion points: Untrusted data enters the agent context through the Case name, Zod snippet, and Expected inputs defined in SKILL.md.\n
Boundary markers: Absent. There are no instructions or delimiters provided to ensure the agent treats the user-provided snippets as non-executable data or to escape potential payload content.\n
Capability inventory: The skill has the capability to write files to the local filesystem (SKILL.md Step 3) and execute shell commands (SKILL.md Step 4).\n
Sanitization: Absent. The skill does not specify any validation or filtering for the user-provided snippets or parameters before they are used in filesystem and shell operations.

schema-fixture