Colin4k1024

上下文优化技能，通过 trigger-based 懒加载和 context composition awareness 自动重组长会话上下文。

Goal-oriented autonomous loop with external completion oracle. Keeps iterating until verifiable stopping conditions are met, checked by a separate model.

Scheduled discovery automation that runs scans on a heartbeat interval, classifies findings, and routes them to goals or triage inbox.

Active MCP connector polling that integrates external tools (issue trackers, error monitoring, CI) into the heartbeat loop for cross-system discovery.

Structured rework with git blame attribution for failed goal iterations. Maps failures to responsible changes and routes targeted fix briefs.

将 CodeGraph 作为默认内置 MCP-backed 代码图谱能力接入，用于 brownfield 符号搜索、 调用链、影响面和上下文查询。输出必须回落到 `/team-*` 主链和 artifacts。

Use Open Design as an external local-first design workbench for prototypes, decks, mobile screens, dashboards, DESIGN.md systems, and exportable visual artifacts. Keep TSP responsible for team workflow, handoff, and quality gates.

将 GitNexus 作为受控可选代码智能能力接入，用于 brownfield MCP 查询、影响面分析、 detect_changes、多仓分析和更深代码图谱证据。输出必须回落到 `/team-*` 主链和 artifacts。

Refines rough ideas through Socratic questioning, explores alternatives, presents design in digestible chunks for validation. Activates before writing any code or making major decisions.

Manages concurrent subagent workflows with task distribution, progress tracking, and result aggregation. Use when multiple independent tasks can run simultaneously.

Executes implementation plans task-by-task with two-stage review (spec compliance, then code quality). Enforces TDD cycle: RED-GREEN-REFACTOR. Activates when user says 'go' or approves plan.

Verifies completed work, runs tests, presents merge/PR options, and cleans up worktrees. Activates when all implementation tasks are complete.

Responds to code review feedback constructively, implements fixes, and communicates progress. Handles feedback professionally and improves code quality.

Reviews code against plan for correctness, bugs, security issues, and adherence to standards. Reports issues by severity. Activates between tasks or before merging.

Dispatches parallel subagents for concurrent task execution with two-stage review (spec compliance + code quality). Activates for large plans (10+ tasks) to maximize throughput.

Applies 4-phase root cause analysis: gather data, form hypothesis, test hypothesis, solve and verify. Includes root-cause-tracing, defense-in-depth, and condition-based-waiting techniques. Activates for any bug.

Eino ADK WrapInvokableToolCall middleware pattern for intercepting all tool calls. TRIGGER when: implementing tool-level middleware (sandbox, logging, auth, metrics) in this project's Eino ADK agents. DO NOT TRIGGER when: working with agent-level middleware (timeout, retry) or non-Eino tool systems.

Reference for the A2UI SSE streaming protocol used by Superagent agents to push typed events to the frontend. TRIGGER when: implementing a new event type, debugging streaming output, writing frontend SSE consumers, or when the user asks "how does streaming work", "what events does the agent emit", "A2UI 协议". DO NOT TRIGGER when: working on HTTP REST endpoints unrelated to streaming.

Author and validate Superagent declarative YAML agent definitions (apiVersion: superagent/v1, kind: Agent). TRIGGER when: creating a new agent, editing configs/agents/*.yaml, reviewing agent spec fields, or when the user asks "how do I define an agent", "write me an agent yaml", "agent 配置怎么写". DO NOT TRIGGER when: working on workflow DAG nodes (use workflow-dag skill), pure Go backend logic, or infrastructure config.

Analyzes code errors, generates fix solutions, evaluates and applies fixes. Use when fixing test failures, compilation errors, runtime exceptions, "编码修复", "code fix", "fix this error", or when the user asks to debug and fix code issues.

Implement and debug agent interrupt/resume (human-in-the-loop checkpointing) in Superagent. TRIGGER when: enabling interrupt on an agent, implementing resume API, debugging "checkpoint not found" errors, adding human approval steps, or when the user asks "interrupt/resume 怎么用", "how does checkpoint work", "pause agent for approval". DO NOT TRIGGER when: building non-interactive batch agents or simple one-shot queries.

Patterns for integrating MCP (Model Context Protocol) servers and tools into Superagent agents. TRIGGER when: wiring an MCP server into an agent YAML, implementing a new MCP client/server, debugging "mcp tool not found" errors, or when the user asks "how do I add an MCP tool", "MCP 怎么接入", "connect filesystem MCP". DO NOT TRIGGER when: writing pure builtin tools (use builtin/<name> directly) or REST API integrations unrelated to MCP.

Guide for configuring and debugging the model routing layer in Superagent. TRIGGER when: adding a new model provider, configuring fallback chains, tuning cost/latency routing strategy, debugging "model not found" errors, or when the user asks "how does model routing work", "模型路由怎么配置", "add a new LLM provider". DO NOT TRIGGER when: writing agent YAML (use agent-yaml-authoring), implementing tool logic, or working on the UI.

Design multi-agent systems using Superagent's supervisor, sequential, and parallel agent types. TRIGGER when: building orchestration agents, routing tasks between sub-agents, designing fan-out/fan-in pipelines, or when the user asks "multi-agent 怎么设计", "supervisor agent", "parallel agents", "agent orchestration". DO NOT TRIGGER when: working on single-model agents (use agent-yaml-authoring) or workflow DAG nodes (use workflow-dag skill).

A test skill for integration testing (tenant A exclusive)

A test skill for integration testing (tenant B exclusive)

Design and execute scientific laboratory experiments with rigorous methodology, controls, and statistical analysis.

Conduct rigorous academic peer review of research papers — evaluate methodology, statistical analysis, and contribution to the field.

Navigate the high seas using stars, compass, and nautical charts. Pirate-style navigation for treasure hunting voyages.

Plan and execute pirate treasure hunts — map reading, crew coordination, X-marks-the-spot strategies, and sea navigation tips.

Analyzes log files, identifies errors, warnings, and patterns

Analyzes PDF documents, extracts text, tables, and generates summaries

Maintain the Oris Rust agent runtime with an issue-driven GitHub workflow. Use when Codex is working in the Oris repository and needs to pick up a GitHub issue, turn the issue into scoped code changes, run Rust validation such as cargo fmt and cargo test, publish the oris-runtime crate to crates.io, close the completed issue, and continue to the next iteration.

Use this skill for requests related to state graphs and the graph API in order to fetch relevant documentation to provide accurate, up-to-date guidance.

Maintain the Adaptive Memory System with an issue-driven GitHub workflow. Use when working in the adaptive-memory-system repository and needs to pick up a GitHub issue, implement code changes, run validation (cargo fmt, clippy, cargo test for Rust; npm lint, tsc for frontend), create commits, push changes, close the completed issue, and continue to the next iteration.

Use this skill when reviewing code for security vulnerabilities, performance issues, or quality problems. Activates for security scans, performance audits, code quality checks, and CI/CD integration tasks. Supports Go, Python, TypeScript, JavaScript, Java, Ruby, Rust, C++, PHP, Swift, and Kotlin.