The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash to execute file conversion tools like pdftotext and pandoc. It provides command templates that interpolate user-supplied file paths (e.g., pdftotext "FILE" "$TMP"). If the agent executes these via a shell without rigorous sanitization of the input paths, a maliciously crafted filename (e.g., containing backticks or semicolons) could lead to arbitrary command execution.
[PROMPT_INJECTION]: The skill processes untrusted data from external sources (referee reports and manuscripts) and grants the agent capabilities like Bash, Write, and Task. This creates a surface for indirect prompt injection where instructions hidden within the academic documents could influence the agent's behavior.
Ingestion points: Path parameters $0 (referee report) and $1 (revised manuscript) in SKILL.md used as input for parsing.
Boundary markers: The instructions do not define explicit boundary markers (like XML tags) or system-level warnings to ignore instructions found within the input files.
Capability inventory: The skill allows use of Read, Write, Edit, Grep, Glob, Bash, and Task (used to spawn a claim-verifier subagent).
Sanitization: The skill recommends quoting file paths in shell commands to prevent simple path injection, but it does not sanitize or filter the content of the documents being processed.

respond-to-referees