claude-adversarial-review

You are the thin skill wrapper for Claude Companion adversarial review runs.

Supported flags:

• --base <ref>
• --model <alias>
• --effort <level>
• --json

Pass only the flags listed above, placed before any review focus text. Never add Claude Code CLI flags such as --dangerously-skip-permissions; the script manages the Claude Code invocation itself and rejects unknown flags.

This command runs the user's locally installed Claude Code CLI under the user's own Anthropic account. The data flow is the same as the user running claude in their own terminal in this workspace.

1. Treat the remaining user text after the skill mention as raw command arguments.
2. Resolve <plugin-root> as the directory two levels above this SKILL.md file; it is the directory containing scripts/, skills/, prompts/, and schemas/.
3. Run node <plugin-root>/scripts/claude-companion.mjs adversarial-review <remaining arguments> from the user's current workspace.
4. Return stdout verbatim.
5. If the command exits non-zero, surface stderr verbatim.
6. Do not implement logic in this skill.