一键导入
rpi
Use when implementing features from Jira tickets, PRDs, or user requirements. Orchestrates Research-Plan-Implement workflow with quality gates.
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
Use when implementing features from Jira tickets, PRDs, or user requirements. Orchestrates Research-Plan-Implement workflow with quality gates.
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。
Validates research or plan against hallucination, overscoping, and traceability. Produces a clear PASS/WARN/FAIL verdict.
Reviews code for correctness, security, performance, and pattern compliance. P0/P1/P2 severity. Absorbs security and performance audit checks.
Executes implementation plan with quality checks and progress tracking. Follows AGENTS.md patterns strictly.
Creates detailed implementation plan from validated research. Produces task breakdown with dependencies and file inventory.
Use when needing to understand requirements before implementation. Gathers context from Jira, Confluence, codebase, and docs. Produces research document with confidence assessment.
基于 SOC 职业分类
| name | rpi |
| description | Use when implementing features from Jira tickets, PRDs, or user requirements. Orchestrates Research-Plan-Implement workflow with quality gates. |
Full workflow orchestrator that invokes individual skills in sequence with quality gates.
Input → Research → Audit → Plan → Audit → Approve → Implement → Code Review
.claude/output for Claude Code, .codex/output for Codex CLI.Derive feature name from input (used for all output files):
KB-1234 → kb-1234export-csv)Resume: If output files for this feature already exist in OUTPUT_DIR (e.g., research-{feature}.md), read them and skip to the appropriate step.
Invoke /research {input}. Produces OUTPUT_DIR/research-{feature}.md.
Invoke /audit research. Quality gate:
| Check | Gate |
|---|---|
| All requirements traceable | No phantom requirements |
| Confidence high enough to plan | No major open unknowns |
FAIL → Stop, report findings, ask user for clarification, re-run /research.
PASS → Proceed.
Invoke /plan. Produces OUTPUT_DIR/plan-{feature}.md.
Invoke /audit plan. Quality gate:
| Check | Gate |
|---|---|
| Every requirement has a task | Full traceability |
| No unnecessary scope added | No overengineering |
| No requirements missed | No underengineering |
| Follows AGENTS.md patterns | Pattern compliance |
FAIL → Revise plan, re-run /audit plan.
PASS → Proceed.
Present a concise plan summary:
Feature: {feature name}
Tasks: {count} | New files: {n} | Modified: {n}
Tasks:
T1: {title}
T2: {title}
...
Quality Gates: Research Audit ✓ | Plan Audit ✓
Proceed with implementation? (yes/no)
Wait for explicit approval. If user requests changes, revise and re-audit before re-presenting.
Invoke /implement. Executes tasks in dependency order, verifies each, runs lint.
Invoke /code-review. Reviews all new and modified files for correctness, security, and performance with P0/P1/P2 findings.
| Severity | Action |
|---|---|
| P0 (Critical) | Must fix before completing |
| P1 (Important) | Should fix, discuss with user |
| P2 (Nice-to-have) | Note for future |
If P0 found: fix all P0 issues, re-run /code-review.
| File | Phase |
|---|---|
research-{feature}.md | Research |
audit-research-{feature}.md | Research Audit |
plan-{feature}.md | Plan |
audit-plan-{feature}.md | Plan Audit |
review-{feature}.md | Code Review |
/rpi KB-1234 # From Jira
/rpi {confluence-url} # From Confluence
/rpi "Add export feature" # From direct prompt
Resume after context loss: Output files in OUTPUT_DIR are the source of truth. Read research-{feature}.md and plan-{feature}.md to resume at the right step.
Individual skills (standalone):
/rpi for complete feature implementation