メニュー
Use when development work has independent lanes for research, codebase exploration, inventory, validation, review, or non-overlapping implementation that can run in subagents, parallel tool calls, or separately owned workstreams.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Use when a development task is unclear, multi-step, risky, or needs coordinated gates for shared understanding, research, parallel lanes, workflow artifacts, and verification. For a single concern, use the matching gate skill if available; for tiny mechanical changes, skip unless the user asks.
Use when development work is large, risky, multi-session, multi-agent, interrupted, or needs durable plan/handoff memory for continuity across coding agents.
Use when development work depends on current or external facts: latest docs/APIs, library or platform behavior, security guidance, unfamiliar errors, migrations, comparisons, best practices, prior art, or known fixes.
Use when a development request is vague, exploratory, conflicted, under-specified, or the user says they are unsure, wants to be grilled, wants assumptions challenged, or needs goal, scope, constraints, and success criteria clarified before planning or implementation.
Use before saying development work is done, or when choosing project-shaped tests/checks to prove code behavior, bug fixes, refactors, config, infrastructure, or UI changes.
Pi-only web research workflow for environments that have pi-9router-ext and a subagent extension such as @tintinweb/pi-subagents configured. Use this skill when the user explicitly asks to use 9router, pi-9router-ext, ninerouter tools, or a 9router route/combo for web research, or when project/user instructions state that 9router-web-researcher is the preferred research stack. Do NOT use for ordinary web research in standard shared-mcp-only or non-Pi clients; use web-researcher instead. This skill always delegates research lanes to subagents and then synthesizes compact cited findings.
| name | parallel-lanes |
| description | Use when development work has independent lanes for research, codebase exploration, inventory, validation, review, or non-overlapping implementation that can run in subagents, parallel tool calls, or separately owned workstreams. |
Use parallelism when independence is real.
Before defining lanes, name the available execution mechanism: subagents, parallel tool calls, or sequential fallback. If the runtime has no subagent or parallel-tool support, collapse the lanes into sequential tasks while keeping the same lane contract and synthesis.
Use the strongest available mechanism:
Define each lane with:
The main agent owns synthesis, decisions, and final user communication.
Default lanes are research, exploration, and validation. Editing lanes are allowed only when file ownership is clearly non-overlapping. If ownership is unclear, lanes report findings and the main agent applies edits.
Every lane has a result, gap, or blocked reason. Before leaving the gate, each gap or blocker is resolved by the main agent, marked non-blocking with reason, added to the plan/artifact as follow-up, or escalated to the user. The main agent has synthesized how the lane output changes the plan.
