一键导入
scale
Run Claude at scale. Non-interactive mode, parallel sessions, fan-out across files, CI integration. Validate on small samples before full runs.
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。
菜单
Run Claude at scale. Non-interactive mode, parallel sessions, fan-out across files, CI integration. Validate on small samples before full runs.
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
Turn ambiguous requirements into structured spec documents. Surface assumptions → interview → spec → human review → execute in fresh session. Spec is a contract, not a suggestion.
Pre-merge review. Reviews the diff, auto-fixes safe issues, flags destructive commands, runs verification. Evidence before assertions.
Execute an implementation plan with verification at every step. Tests first, then code. Stop on failures — never stack problems.
Systematic debugging. Reproduce → isolate → root cause → fix → verify. Root cause confirmed before any fix is applied. Never treat symptoms.
Build frontend interfaces with a committed aesthetic direction. Screenshot-based self-verification loop. Not generic defaults.
Read-only codebase exploration. Maps structure, patterns, and boundaries before any code is written. Use subagents to isolate context.
| name | scale |
| description | Run Claude at scale. Non-interactive mode, parallel sessions, fan-out across files, CI integration. Validate on small samples before full runs. |
Move beyond one-human-one-Claude. Run non-interactive, parallel, and in CI.
claude -p "Explain what this project does"
claude -p "List all API endpoints" --output-format json
claude -p "Analyze this log file" --output-format stream-json
Use in: CI pipelines, pre-commit hooks, scripts, data pipelines.
For large migrations or batch operations:
for file in $(cat files.txt); do
claude -p "Migrate $file: [specific instructions]. Return OK or FAIL." \
--allowedTools "Edit,Bash(git commit *)"
done
--allowedTools to scope permissions for unattended runs.| Pattern | How |
|---|---|
| Writer/Reviewer | Session A implements, Session B reviews (clean context = unbiased review) |
| Test-first | Session A writes tests, Session B writes code to pass them |
| Multi-feature | Independent features in parallel sessions with isolated worktrees |
claude --permission-mode auto -p "fix all lint errors"
Classifier reviews commands before execution. Blocks scope escalation, unknown infrastructure, hostile actions. For -p runs, aborts if repeatedly blocked (no user to fall back to).
claude -p "<prompt>" --output-format json | your_command
cat error.log | claude -p "Summarize the errors"
--allowedTools to restrict what batch runs can do.--verbose during development, turn off in production."Once you're effective with one Claude, multiply your output with parallel sessions, non-interactive mode, and fan-out patterns."