execute
Orchestrator that runs the full APEX workflow (research → plan → implement → ship) in a single session. Use for tasks you want to complete without context switches.
SKILL.md
Full skill instructions
<skill name="apex:execute" phase="orchestrator">
<overview>
Meta-skill that chains all 4 APEX phases in sequence:
/apex:research- Intelligence gathering/apex:plan- Architecture design/apex:implement- Build and validate/apex:ship- Review and reflect
Use this for single-session task completion. For multi-session work, invoke individual skills. </overview>
<when-to-use> - Small to medium tasks that fit in one session - When you don't need to pause between phases - When you want full workflow without manual skill invocation </when-to-use> <when-not-to-use> - Large complex tasks that need human review between phases - When you want to pause after research to think - When context overflow is likely - When you want to run phases in separate sessions </when-not-to-use> <initial-response> <if-no-arguments> I'll run the full APEX workflow. Please provide: - Task description (e.g., "implement dark mode toggle") - Ticket ID (e.g., "APE-59") - Path to task fileExample: /apex:execute "add user authentication"
</if-no-arguments>
<if-arguments>Begin full workflow.</if-arguments>
</initial-response>
Task: [Title] File: ./.apex/tasks/[identifier].md
Phases Completed:
- ✅ Research - [summary]
- ✅ Plan - [chosen architecture]
- ✅ Implement - [files changed, tests status]
- ✅ Ship - [commit SHA, reflection status]
Metrics:
- Total patterns applied: [N]
- Tests: [passed]/[total]
- Review findings: [N] ([dismissed]% false positives)
- Commit: [SHA]
Key Learning:
[From reflection]
Task complete. Full history in ./.apex/tasks/[identifier].md
</template>
</step>
</workflow>
<error-handling> <phase-failure> If any phase fails: 1. Report which phase failed 2. Report current state of task file 3. Suggest manual intervention: "Run `/apex:[phase] [identifier]` to retry" 4. Do NOT continue to next phase </phase-failure> <user-input-needed> If any phase needs user input: 1. Present the question 2. Wait for response 3. Continue that phase with response 4. Do NOT skip the interaction </user-input-needed> <context-overflow> If context is getting large: 1. Warn user: "Context is large. Consider continuing in new session." 2. Report current phase and identifier 3. User can restart with individual skill from current phase </context-overflow></error-handling>
<success-criteria> - All 4 phases completed successfully - Task file shows phase: complete, status: complete - Git commit exists - Reflection recorded - Final report displayed to user </success-criteria></skill>
