pi-lot

Use this for LlmAgent work: agent loop changes, AgentSession/harness APIs, AgentLoopOptions/API design, tools, events, context budget, option forwarding, or any task that mentions the agent package. It forces a harness-first public-surface decision, prefers portable SDK context streaming for ergonomic agent behavior, keeps raw Responses as an explicit escape hatch, and picks unit/fake/live agent tests before implementation.

Batch workflow for llm-svc issue slates. Use this when the user asks to work through a group of GitHub issues by label, priority, milestone, or explicit issue list. It triages the slate, groups issues into small independent PRs, creates a roadmap and todos, then executes each item with a TDD mini-plan.

Use this when adding or reviewing fake/live integration tests in this repo, especially for SDK, service, or agent behavior. It creates paired .Int coverage: fake tests for deterministic happy, negative, edge, and error cases; live Smoke tests for small real-upstream happy-path compatibility.

Use this for LlmSdk work: SDK issues, SDK client changes, Core/Proxy behavior, Infrastructure adapters, model/auth/routing/translation changes, or any task that mentions the SDK layer. It forces an SDK layer decision first, keeps agent/service consumers separate, and picks the owning SDK tests before implementation.

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.