aws-samples / aws-samples/sample-autonomous-cloud-coding-agents
feat(agent): tiered validation pipeline (Tiers 2–3)
- Lingua principale
- TypeScript
- Stelle
- 143
- Fork
- 46
- Merge medio
- 3g 9h
- PR unite (30g)
- 20
Descrizione
**Context:** ROADMAP.md → Agent quality → Tiered validation pipeline
**Related:** #149 (ADR-013 closed), in-pipeline fix-up draft, evaluation pipeline draft
---
## Component
Agent (Python runtime)
## Describe the feature
Three post-agent tiers per roadmap:
1. **Tier 1 (already shipped):** tool validation—build/test/lint via agent post-hooks and PR body baselines. *This issue implements Tiers 2–3 only.*
2. **Tier 2:** code quality—DRY/SOLID/complexity analysis.
3. **Tier 3:** risk and blast radius analysis—feeds PR risk classifier and review routing (#425).
See ADR-013 tiered validation pyramid.
## Use case
Build/lint passing does not imply maintainable or low-risk changes. Operators want graduated validation before PR creation and to feed evaluation metrics.
## Proposed solution
1. Define tier interfaces in `agent/` with Blueprint opt-in per repo.
2. Tier 2: integrate existing linters/analyzers (eslint complexity, ruff, optional semgrep rules) with structured JSON output.
3. Tier 3: rule-based blast radius (files touched, public API surface, migrations)—output consumed by PR risk classifier draft.
4. Non-blocking advisory mode first; optional Cedar/policy gates later.
5. Emit validation results to `TaskEventsTable` and task record for dashboards.
## Other information
- ADR-013 documents the validation pyramid; this issue tracks **implementation**.
- Design context: `docs/decisions/ADR-013-tiered-validation-pyramid.md`, `docs/design/EVALUATION.md`.
- [ ] This might be a breaking change
Guida per i contributori
Apri la guida per i contributori
Direzione di ricerca
Inizia leggendo docs/decisions/ADR-013-tiered-validation-pyramid.md e docs/design/EVALUATION.md, quindi esamina le interfacce di agent/ e i riferimenti a TaskEventsTable. Definisci come Tiers 2–3 si integrano con Blueprint opt-in, i risultati consultivi strutturati, l’output di blast-radius e i record delle attività; il lavoro è completato quando questi risultati possono alimentare i dashboard e il classificatore del rischio di PR.
Scritto dal modello di indicizzazione a partire dal testo della issue.
Valutazione
- Stack tecnologico
- python
- Ambito
- ai, devtools
- Tipo di issue
- Funzionalità
- Difficoltà
- 5/5
- Tempo stimato
- Più di una settimana
- Stato di attività
- Tranquilla
- Chiarezza
- Abbastanza chiara
- Idoneità per principianti
- 35/100