The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external documents (Chinese source and English translation) provided by the user. These documents could contain malicious instructions or hidden prompts that attempt to influence the agent's behavior or bias the translation report.
Ingestion points: File content is read via pandoc or direct reading of .md/.txt files as specified in Step 2 of the workflow.
Boundary markers: The instructions do not define clear delimiters or "ignore previous instructions" guards when interpolating document content into the review prompts.
Capability inventory: The agent has the capability to execute shell commands (pandoc), interact with the user via AskUserQuestion, and write files to the local file system.
Sanitization: No sanitization or filtering logic is specified for the content retrieved from the processed documents.
[COMMAND_EXECUTION]: The workflow in SKILL.md (Step 2) instructs the agent to execute pandoc shell commands using filenames provided by the user. While the filenames are wrapped in double quotes in the template, a maliciously crafted filename (e.g., containing backticks or subshell markers like $(...)) could potentially bypass this if the execution environment does not properly sanitize inputs, leading to arbitrary command execution.

lov-translation-review