The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs and executes multiple shell commands (including git merge, git tag, git branch -D, and python script invocations) by directly interpolating variables such as {session-id}, {winner}, and {base_branch}. If these variables are sourced from untrusted data (e.g., attacker-controlled agent names or session metadata) and are not properly sanitized or escaped, an attacker could execute arbitrary shell commands.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability surface as it processes untrusted data from a preceding /hub:eval session to determine execution logic.
Ingestion points: Data enters the context from previous session results or user-supplied arguments for --agent and session IDs.
Boundary markers: No boundary markers or 'ignore' instructions are used when interpolating these variables into shell commands.
Capability inventory: The skill has the ability to delete git branches (git branch -D), modify repository tags, and execute local Python scripts (session_manager.py).
Sanitization: No validation, sanitization, or escaping of the input variables is present in the provided instructions.

merge