The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates the execution of several external agent CLIs, including claude, codex, agy, and grok. These are invoked as headless subagents to perform parallel task processing.
[PROMPT_INJECTION]: The skill provides explicit instructions to prevent shell command injection when handling user-supplied task strings. By mandating the use of quoted heredocs (cat <<'PANEL_EOF') and quoted variable expansion ("$TASK"), it ensures that the shell does not interpret special characters (like backticks or dollar signs) contained within the user input.
[COMMAND_EXECUTION]: Automated execution is enabled using bypass flags such as --dangerously-skip-permissions and --always-approve. To mitigate the risk of unintended side effects, the skill includes a mandatory verification step using git status --porcelain after subagent execution to identify and revert any unauthorized file modifications.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from user prompts or git diffs.
Ingestion points: User-provided task strings and git diff output (SKILL.md Step 1).
Boundary markers: Use of quoted heredoc delimiters to isolate the task string during shell interpolation (SKILL.md Step 3).
Capability inventory: Execution of external agent CLIs with auto-approval flags enabled (SKILL.md Step 3).
Sanitization: Automatic reversion of any file changes using git checkout after tree status verification (SKILL.md Step 3).

panel