aws-samples / aws-samples/sample-autonomous-cloud-coding-agents
feat(agent): PR scope creep check in pr_review workflow
- Langage dominant
- TypeScript
- Étoiles
- 143
- Forks
- 46
- Merge moyen
- 3 j 10 h
- PR mergées (30 j)
- 24
Description
**Context:** ROADMAP.md → Agent quality → PR scope creep check (`pr_review`)
**Related:** pr-risk-classifier draft, #425
---
## Component
Agent (Python runtime)
## Describe the feature
Advisory-first scope analysis in `coding/pr-review-v1` comparing **declared intent** (task description / issue / PR narrative) to the **actual diff** and touched areas.
Structured output:
| Field | Values / content |
|-------|------------------|
| `scope_rating` | `within_scope`, `mild_expansion`, `significant_expansion`, `likely_scope_creep` |
| `confidence` | numeric or enum |
| `rationale` | files touched, API/schema/config changes, unrelated dependency churn |
**Rollout:** non-blocking reviewer guidance first; optional policy gates for high-risk repos later.
## Use case
Agents often expand beyond the stated task. Reviewers need early signal before deep review. Teams want consistent scope discipline without blocking low-risk expansions by default.
## Proposed solution
1. Extend `pr_review` workflow step to invoke scope analysis after diff is available.
2. Post structured comment via GitHub Reviews API (advisory section).
3. Emit `scope_analysis` event to `TaskEventsTable`.
4. Unit tests with golden fixtures for rating boundaries.
5. Document prompt and output schema in `docs/design/WORKFLOWS.md`.
## Other information
- Design context: `docs/design/EVALUATION.md`, `agent/workflows/`.
- [ ] This might be a breaking change
Guide de contribution
Ouvrir le guide de contribution
Piste de recherche
Commencez par le workflow `pr_review` dans `agent/workflows/` et le contexte de conception dans `docs/design/EVALUATION.md` et `docs/design/WORKFLOWS.md` ; examinez comment le diff devient disponible et comment les événements de `TaskEventsTable` sont émis. Le travail est terminé lorsque des évaluations de portée, une sortie de revue consultative, un événement `scope_analysis`, des fixtures de limites dorées et une prise en charge documentée de prompt/schema prennent en charge le flux proposé.
Rédigé par le modèle d'indexation à partir du texte de l'issue.
Évaluation
- Stack technique
- github, python
- Domaine
- api, backend, documentation, testing
- Type d'issue
- Fonctionnalité
- Difficulté
- 4/5
- Temps estimé
- 3-5 jours
- Activité
- Calme
- Clarté
- Plutôt claire
- Accessibilité débutants
- 42/100