The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from a user-provided absolute file path and passes it to an LLM oracle for quality evaluation without sufficient boundary markers or sanitization. Malicious instructions embedded in the target file could influence the oracle's output.
Ingestion points: The skill reads the content of the file specified by the target_file parameter and includes it in the prompt for the oracle subagent in section §3.
Boundary markers: The prompt provided to the oracle subagent does not utilize delimiters or specific instructions to disregard potentially malicious commands embedded within the document being reviewed.
Capability inventory: The skill is authorized to use Read and Edit tools for file manipulation, as well as a bash tool for command execution.
Sanitization: No sanitization or validation of the file content is performed before it is analyzed by the subagent.
[COMMAND_EXECUTION]: The skill automatically executes local shell commands via bash to validate changes after modifying a file. This creates a risk where a compromised oracle recommends an edit that introduces malicious code into a script or configuration file, which is then executed during the validation step.
The skill specifies the execution of bun test and openspec schema validate inside a bash shell in section §4.2.
The execution occurs automatically as part of the iterative 'repair-and-verify' loop.

score-prompt