Kiro → Cursor → Verify
Spec in Kiro, implement with Cursor, validate with project commands.
Kiro → Cursor → Verify
This handoff aligns spec-doc §11.2 with AgentFlow gates: authoring stays in Kiro (.kiro/specs/<feature>/), implementation happens wherever Cursor (or cursor-agent) excels, and the CLI enforces deterministic verify/review steps so automation cannot silently diverge.
When to use
You keep requirements and tasks in Kiro (.kiro/specs/<feature>/) but want Cursor (or cursor-agent) for implementation, with mandatory verify/review steps.
Pipeline
The sequence below is illustrative; downstream reporting still consumes persisted run identifiers.
flowchart LR
S[spec] --> P[plan]
P --> E[enrich]
E --> D[dev]
D --> V[verify]
V --> R[review]
R --> REP[report]Commands
Replace billing-v2 with your feature id:
agentflow spec billing-v2 --agent kiro
agentflow plan billing-v2
agentflow enrich billing-v2 --agent ollama
agentflow dev billing-v2 --agent cursor
agentflow verify billing-v2
agentflow review billing-v2 --agent codex
agentflow report <run-id>Use --dry-run on any step while rehearsing:
agentflow dev billing-v2 --agent cursor --dry-runConfiguration defaults
From .agentflow/config.yaml.example:
work:
default_agent: cursor
default_reviewer: codex
default_enricher: ollama
auto_verify: true
auto_review: falseSet work.auto_review: true only when you want review after every successful verify.
Intent shortcut
agentflow work "develop billing-v2" --stop-after verifyIntent resolution picks the feature; V3 pipeline applies budgets and context optimization.
Failure modes
| Symptom | Fix |
|---|---|
kiro not on PATH | Set agents.kiro.command or install Kiro CLI |
| Verify fails | Fix tests locally; use agentflow verify billing-v2 --force only when state machine allows |
| Dirty git blocked | Commit/stash or adjust policies.require_clean_git |