菜单
Delegate self-contained research, review, planning, debugging, or implementation subtasks to a Pi subagent running the openai-codex/gpt-5.5 model. Use when the user asks for Pi, GPT-5.5, a subagent, independent analysis, parallel investigation, or a second coding-agent opinion.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Interview Matt about the next product/dev iteration, turn the discussion into a focused iteration plan, publish planning artifacts, validate the plan, and ask whether this LLM session should launch the Fabro implementation-and-review delivery workflow. Use when planning the next iteration, shaping work before implementation, or preparing a plan for delivery.
Use when writing or reviewing Gherkin scenarios, especially after discovering examples or edge cases that reveal a business rule
Offer to capture and, when accepted, write a docs/kaizen observation when we notice friction, imperfections, waste, opacity, brittleness, or failure in the pipeline/factory/workflows/tooling used to create the product. Use for workflow/tooling problems, not ordinary product bugs.
Read, audit, and safely manipulate Resend email configuration and message data through the Resend API. Use when inspecting domains, API keys, audiences, contacts, broadcasts, emails, webhooks, or making explicit Resend changes with an API key from the environment.
Read, audit, and safely manipulate Postmark configuration and message data through the Postmark API. Use when inspecting servers, streams, senders, domains, templates, messages, webhooks, suppressions, or making explicit Postmark changes with API tokens from the environment.
Read, audit, and safely manipulate DNSimple DNS configuration through the DNSimple API. Use when inspecting domains, zones, DNS records, or making explicit DNS changes with a DNSimple API token from the environment.
| name | pi-subagent-gpt55 |
| description | Delegate self-contained research, review, planning, debugging, or implementation subtasks to a Pi subagent running the openai-codex/gpt-5.5 model. Use when the user asks for Pi, GPT-5.5, a subagent, independent analysis, parallel investigation, or a second coding-agent opinion. |
| user-invocable | true |
Use this skill to delegate a bounded task to a separate Pi coding-agent process using Matt's openai-codex/gpt-5.5 model.
From the repository root, run the helper script with a heredoc prompt:
.claude/skills/pi-subagent-gpt55/scripts/delegate-pi-gpt55.sh --research <<'PROMPT'
Your delegated task here.
PROMPT
For tasks that may edit files, use --write:
.claude/skills/pi-subagent-gpt55/scripts/delegate-pi-gpt55.sh --write <<'PROMPT'
Your delegated implementation task here.
PROMPT
The helper invokes:
pi --provider openai-codex --model gpt-5.5 --thinking high -p ...
Delegate when a task is useful as an isolated branch of work:
Do not delegate tiny lookups you can do faster yourself, ambiguous tasks that need user clarification first, or broad implementation work that would conflict with your own edits.
Write the subagent prompt as a complete brief. Include:
dev check.--research)Use for investigation, review, planning, and second opinions. The helper restricts Pi to read/search tools so the subagent cannot edit files. Ask for concise findings with file references.
--write)Use only when you want Pi to change files. Make the scope narrow and avoid editing the same files yourself while the subagent runs. Tell the subagent to run the smallest relevant checks; for Memba code/config/behaviour changes, tell it to run dev check before reporting back unless you deliberately want only a draft patch.
You may run multiple Pi subagents in parallel only when their tasks are independent and either read-only or assigned non-overlapping write scopes. Use separate Bash tool calls in the same turn.