ワンクリックで
architect-interview
Use this skill when in plan mode. It ensures all assumptions are resolved through an interview process.
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。
メニュー
Use this skill when in plan mode. It ensures all assumptions are resolved through an interview process.
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 before shipping high-stakes output (pre-tapeout RTL, verification infra, production scripts) that needs independent adversarial review. Requires reviewer backends configured in santa-method.json — if none are configured, do not use.
Use before shipping high-stakes output (pre-tapeout RTL, verification infra, production scripts) that needs independent adversarial review. Requires reviewer backends configured in santa-method.json — if none are configured, do not use.
Use when context is getting heavy, at natural breakpoints between subtasks, or when the user asks to checkpoint — writes the agent's full working context to a recoverable file before compaction
Audit the Obsidian vault for organizational health — broken links, naming violations, misplaced notes, sync status, stale content, and connection quality
Use when taking notes, recalling context, building connections, or treating the Obsidian vault as Claude's persistent memory across sessions
Required rules for working with the Obsidian vault — tool selection decision tree, quality gates, frontmatter schema, filename conventions, archival, content integrity. Load whenever vault notes are read, written, or referenced.
| name | architect-interview |
| description | Use this skill when in plan mode. It ensures all assumptions are resolved through an interview process. |
| agent | Plan |
| user-invocable | false |
Activate this skill strictly whenever the user requests the creation of new code, system architecture, hardware/software design, or technical project scaffolding.
Treat the requested architecture as a Directed Acyclic Graph (DAG) of dependencies.
Your primary job in this mode is an investigator, not a typist. You must interview the user relentlessly until a 100% shared understanding is reached.
Never ask a purely open-ended question. Every time you ask the user for a decision, you must provide a strong, expert recommendation based on industry best practices, along with a brief rationale (trade-offs).
When triggered, execute the following loop. Do not batch more than 1-2 critical questions at a time.
Step 0: Consult the Vault
Before the first question, check prior art: vault_project(name=<project>) and vault_semantic_search for related type: decision notes. A previously approved decision resolves a node without a question — cite it instead of re-asking; only re-open it if the new request conflicts with it.
Step 1: Map the Immediate Dependencies Identify the most critical, unresolved architectural decision required to move forward.
Step 2: Prompt the User Present the decision to the user using the following strict format:
Step 3: Halt and Await Input Stop generating immediately after Step 2. Do not write the code. Wait for the user's response.
Step 4: Branch Resolution
Step 5: Final Consensus
Once all branches are resolved, output a comprehensive [Architecture Summary]. Only exit "planning mode" and begin generating the actual code/artifacts once the user explicitly approves this summary.