菜单
Analyze tasks upfront before execution. Predict task category, identify key files, assess risk level, and detect high-consequence operations. Use proactively when any task description is provided to guide execution strategy.
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。
Combine findings from parallel intelligence gathering into unified execution plan. Synthesize predictions, web research, deep strategy, and environment observations into optimal approach. Use when multiple intelligence sources need integration before execution.
Generate comprehensive multi-approach strategies before execution. Extract domain knowledge, create alternative approaches, identify failure modes, and develop risk-aware plans. Use proactively for complex tasks requiring strategic thinking or when multiple approaches might succeed.
Perform heuristic environment discovery beyond basic directory listing. Check installed packages, analyze folder structure, inspect running processes, and assess system state. Use proactively when starting new tasks or when environment context is unclear.
Handle common execution failures with specialized recovery strategies. Fix syntax errors, import/dependency issues, path/file problems, permission denial, and connection timeouts. Use proactively when encountering errors or as automatic recovery mechanism.
Prevent premature task completion by checking for parsing errors, execution failures, incomplete test results, and missing expected outputs. Acts as final quality gate before declaring tasks complete. Use proactively after major operations or when task success needs verification.
Multi-round web search with low-frequency query generation. Prioritize GitHub/StackOverflow for actionable code, extract AI Overview insights, and perform deep link analysis. Use proactively for complex problems requiring external knowledge or when stuck on implementation challenges.
| name | predictive-intelligence |
| description | Analyze tasks upfront before execution. Predict task category, identify key files, assess risk level, and detect high-consequence operations. Use proactively when any task description is provided to guide execution strategy. |
| allowed-tools | Read, Grep, Glob |
Apex2's prediction phase implemented as a Claude Code skill. This skill extracts critical task metadata before any execution begins.
When invoked (either explicitly or automatically by Claude), perform these sequential analyses:
Analyze the task description to determine:
Extract specific file/folder mentions from the task:
Identify potential high-consequence operations:
rm -rf or destructive commandsPredict what needs to be present:
Based on analysis, provide strategic recommendations:
Provide a concise prediction report with:
Task Analysis:
Category: [category]
Complexity: [level]
Duration: [estimate]
Risk: [profile]
Key Files:
- [file1]: [purpose]
- [file2]: [purpose]
Risks Identified:
- [risk1]: [mitigation]
- [risk2]: [mitigation]
Recommended Strategy:
1. [step1]
2. [step2]
3. [step3]
Environment Needs:
- [requirement1]
- [requirement2]
This skill is automatically useful when:
The goal is to prevent wasted effort and ensure safe, efficient execution by understanding the full scope before beginning.