forge-reviewlisted

# forge-review Independent reviewers, one per dimension (correctness, security, performance, conventions), run in bounded parallel over the same change and their findings are merged into a single deduplicated, severity-sorted report. This is the first consumer of the orchestration primitive (`scripts/_orchestrate.py`). ## When to Use - User runs `/forge:review` - User asks to review the current diff, a file, or the branch - User wants a broad review ("find anything wrong", "is this safe/fast/clean?") ## Pre-flight Background capability is **not** required — review is synchronous orchestration, not a daemon. But it does spend (one bounded agent per dimension), so it is cost-gated through `_cost_cap`. If `.forge/capabilities.json` reports background unavailable, the dimensions still run synchronously via `claude -p`; if the cost cap is hit, dimensions are skipped and noted (no silent truncation). ## How to run Pick the target — the staged diff is the common case: ```bash git diff --staged > /tmp/forge-review-target.diff python3 ${CLAUDE_PLUGIN_ROOT}/scripts/review_synthesize.py \ --target /tmp/forge-review-target.diff --cwd . ``` Or review a specific file, or pipe a diff on stdin (`--target -`): ```bash git diff main... | python3 ${CLAUDE_PLUGIN_ROOT}/scripts/review_synthesize.py --target - --cwd . ``` The script fans the four dimensions out in parallel (deterministic, index-ordered), validates each reviewer's structured findings, drops any malformed dimension with