RPI - Research, Plan, Implement (Orchestrator) Skill

RPI - Research, Plan, Implement (Orchestrator)

Full workflow orchestrator that invokes individual skills in sequence with quality gates.

Input → Research → Audit → Plan → Audit → Approve → Implement → Code Review

Agent Compatibility

OUTPUT_DIR: .claude/output for Claude Code, .codex/output for Codex CLI.
Invoke sub-skills with the Skill tool (Claude Code) or by naming the skill in the prompt (Codex CLI / Copilot CLI).

When to Use

User provides a Jira issue key (e.g., KB-1234)
User provides a Confluence PRD URL
User describes a feature to implement

Step 1: Input & Feature Naming

Derive feature name from input (used for all output files):

Jira key KB-1234 → kb-1234
Confluence URL → sanitized page title slug
Direct prompt → short slug, max 30 chars (e.g., export-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.

Step 2: Research

Invoke /research {input}. Produces OUTPUT_DIR/research-{feature}.md.

Step 3: Research Audit

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.

Step 4: Plan

Invoke /plan. Produces OUTPUT_DIR/plan-{feature}.md.

Step 5: Plan Audit

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.

Step 6: User Approval

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.

Step 7: Implement

Invoke /implement. Executes tasks in dependency order, verifies each, runs lint.

Step 8: Code Review

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.

Output Files

| 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 |

Quick Reference

/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):

Use /rpi for complete feature implementation
Use individual skills for targeted tasks or exploration

Agent Skills: RPI - Research, Plan, Implement (Orchestrator)

Install this agent skill to your local

Skill Files