// Mandatory pre-commit code review gate. Checks constraint violations, cross-module consistency, and implementation quality. Trigger proactively when changes span multiple files or touch shared infrastructure. Trigger: 'review', 'check before commit'.
| name | ff-review |
| description | Mandatory pre-commit code review gate. Checks constraint violations, cross-module consistency, and implementation quality. Trigger proactively when changes span multiple files or touch shared infrastructure. Trigger: 'review', 'check before commit'. |
1. Capture changes → git diff
2. Load constraints → .agents/knowledge/constraints.md
3. Review against constraints and architecture
4. Route by verdict:
✓ Safe → Proceed with commit
⚠ Needs-attention → Fix issues, then commit
✗ Risky → Halt and report
git diff HEAD # All changes
git status # Modified files
.agents/knowledge/constraints.md — All hard constraints.agents/knowledge/architecture.md — Module dependenciesabc.py base classes reflected in ALL subclasses (GRPO, GRPOGuard, DPO, NFT, AWM)hparams/ reflected in ALL example configsAdvantageProcessor compatible with all trainers_shared_fields consistentself.device or accelerator.device)@torch.no_grad() on reward model __call__line-length=100)profile="black")hparams)try/except ImportError, backend-gated runtime feature checks like DeepSpeed/FSDP, unresolvable circular imports).guidance/ docs updated if behavior changed# Options: comments[{modules}] {type}: {description}No issues found. Proceed with commit.
Issues found but fixable:
Potential breaking changes:
black --check src/ && isort --check src/ to confirm formatting compliance.[{modules}] {type}: {description} format.topics/fix_patterns.md archival process.Before reviewing, always read Tier 1: constraints.md, architecture.md, philosophy.md.
Additionally, read based on diff scope:
| Diff touches... | Also read |
|---|---|
models/ | topics/adapter_conventions.md, topics/parity_testing.md |
trainers/ | topics/train_inference_consistency.md |
scheduler/ | topics/train_inference_consistency.md, topics/dtype_precision.md |
| New adapter | topics/adapter_conventions.md, topics/parity_testing.md |
| dtype/precision | topics/dtype_precision.md |
# Options: commentwait_for_everyone() — Distributed deadlock riskComplete workflow for adding a new RL training algorithm. Covers paradigm selection, TrainingArguments subclass, trainer implementation, registry, example config, and verification. Trigger: 'add algorithm', 'new trainer', 'new training method', 'implement algorithm'.
Complete workflow for adding a new model adapter. Covers analysis, sample dataclass, adapter implementation (4 abstract methods + per-modality encoder overrides), registry, example YAML, and verification. Trigger: 'add model', 'support new model', 'integrate model', 'new adapter'.
Feature development with cross-module impact analysis. Covers trainer hierarchy, model adapters, reward pipeline, config system, sample dataclasses, and distributed training paths. Trigger: 'add feature', 'implement', 'refactor', 'reorganize', 'new capability'.
Bug fixing and debugging for ANY error, crash, loss divergence, gradient explosion, distributed hang, NaN, or unexpected behavior. Covers quick fixes and full protocol with 5-phase investigation. Trigger: 'fix bug', 'fix error', 'broken', 'crash', 'doesn't work', 'fails with', 'loss NaN', 'training hangs', 'OOM'.
Complete workflow for adding a new reward model. Covers pointwise vs groupwise design, __call__ contract, registration, YAML config, multi-reward setup, and verification. Trigger: 'add reward', 'new reward model', 'custom reward', 'scoring function'.