| name | adversarial-review |
| description | Launch parallel red-team and blue-team claude -p sessions to stress-test a manuscript's arguments against opposition papers. |
| argument-hint | <manuscript-path> <papers-dir> [output-dir] [phases] [--excel <path>] |
| allowed-tools | Bash |
Adversarial Review
Run the pipeline script with the user's arguments:
~/.claude/skills/adversarial-review/run.sh "$0" "$1" "$2" "$3"
$0: Manuscript path (absolute)
$1: Papers directory (absolute)
$2: Output directory (default: ./adversarial-review-output)
$3: Phases to run (default: auto-detect missing). Examples: all, 0,1, 2+, 3a, 3b, 3c, 3d, 3e, 3f, 3g, 4
--excel <path>: Optional Excel file with paper abstracts for triage (skips PDF reading)
The script handles all orchestration: template filling, agent launching, parallelism, section extraction, and phase sequencing. It will print progress to stdout.
After the script completes, read and present FINAL-REPORT.md to the user.