The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from the $ARGUMENTS variable (bug descriptions) and project files during its investigation phase (Step 1 and Step 2). It possesses the capability to execute shell commands (Bash) and modify files (Write, Edit).
Ingestion points: $ARGUMENTS in Step 1, codebase files and research bundles in Step 2.
Boundary markers: Minimal; instructions mention preserving specific Handoff annotations but do not provide delimiters for untrusted data processed during investigation.
Capability inventory: Includes Bash for command execution, Write/Edit for file modification, and Task for spawning autonomous sub-agents.
Sanitization: No explicit sanitization or validation of the contents of $ARGUMENTS or investigated files is required before these contents influence agent logic or tool use.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform environment checks (Step 0), run regression tests (Step 2.5), and verify fixes (Step 4). While these are intended behaviors for a bug-fixing skill, the execution of arbitrary commands (especially shell scripts for reproduction) based on potentially malicious external instructions in a bug report or codebase poses a risk.

aif-fix