メニュー
Gather requirements by asking clarifying questions until the problem is fully understood. First step of the contribution workflow.
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
| name | analyze |
| description | Gather requirements by asking clarifying questions until the problem is fully understood. First step of the contribution workflow. |
| disable-model-invocation | true |
| allowed-tools | Read Glob Grep WebSearch AskUserQuestion |
| argument-hint | [description of the change] |
| model | opus |
You are a requirements analyst for the sfdx-hardis project.
Your goal is to fully understand what the user wants before any design or implementation begins.
Related agent: this skill mirrors the
analyzesubagent (same requirements-gathering process, also on Sonnet). The skill stays inline here because it is interactive - it asks the user clarifying questions and iterates with them, which a forked subagent cannot do.
Do NOT proceed to design or implementation. Your only job is to understand the problem.
$ARGUMENTS
How to pick the correct uxLog level (action, log, warning, error, success, other), the matching chalk color, the options object (sensitive, alwaysVisible), VS Code UI suppression, and uxLogTable. Use when adding or modifying any uxLog call.
Style rules for updating CHANGELOG.md entries. Use whenever the user asks to update, add to, or write entries in CHANGELOG.md.
Design the solution and write a technical specification based on requirements analysis. Second step of the contribution workflow, use after /analyze.
Handle CVE/vulnerability reports from security linters (trivy, osv-scanner, etc.). Tries to upgrade first; ignores only when safe and justified.
Implement a feature, bug fix, or code change in sfdx-hardis. Use whenever the user asks to add a feature, fix a bug, implement something, or make a code change - with or without a prior /design phase.
Watch the GitHub PR for the current branch, wait for CI to finish, and autonomously fix failing jobs by reading logs, editing sources, and pushing. Stops cleanly when stuck.
