Pipeline

---
description: Full Claudito pipeline - the complete chain for bulletproof fixes
allowed-tools: Skill, Bash, Read, Grep, Edit, Write
---

# Pipeline - Master Chain Command

## CRITICAL: USER APPROVAL GATE

**NO CODE CHANGES WITHOUT TREY'S EXPLICIT APPROVAL**

The pipeline ALWAYS stops after ARCHITECT and BEFORE FIXER to:
1. REPORT all findings
2. SHOW proposed changes with code preview
3. WAIT for "OK" or "approved" from Trey
4. ONLY THEN proceed to apply fixes

WARDEN → ENTOMOLOGIST → FORENSICS → ARCHITECT ↓ 🛑 USER APPROVAL (Pipeline PAUSES here) ↓ (After Trey approves) ↓ FIXER → DEBUGGER → etc.

## PROMPT CHAINING EXECUTION

This pipeline uses **prompt chaining** - each step outputs in XML tags that feed into the next step.

### STEP 1: WARDEN (Scope Check)

Input: $ARGUMENTS (the task) Output: true/false what's in/out of scope

**IF scope_approved=false, STOP and ask user.**

### STEP 2: FORENSICS (Find Issues)

Input: $ARGUMENTS Output: file:line what's wrong minimal change needed

### STEP 3: ARCHITECT (Plan Fix)

Input: {{from step 2}} Output: how to fix files to modify what NOT to touch

### STEP 4: FIXER (Apply Fix)

Input: {{from step 3}} Output: old code new code

### STEP 5: DEBUGGER (Test)

Input: {{from step 4}} Output: pass/fail pass/fail pass/fail

### STEP 6: CRITIC (Review) - LOOP POINT

Input: {{from step 4}} {{from step 5}} Output: true/false list of issues additional requirements for re-fix

**IF approved=false, LOOP back to FIXER with constraints. Max 3 loops.**

### STEP 7: VALIDATOR (Quality Gate)

Input: {{final}} {{final}} Output: true/false any showstoppers

### STEP 8: FORENSICS AGAIN (Landmine Check)

Input: {{final}} Output: true/false any new bugs introduced

**IF clean=false, mini fix cycle: FIXER → DEBUGGER → VALIDATOR → back here**

### STEP 9: SCRIBE (Document)

Input: All previous outputs Output: Updates to ogz-meta/claudito_context.md

### STEP 10: COMMITTER (Git)

Input: {{final}} Output: abc123 commit message

### STEP 11: CHANGELOG

Input: {{final}} {{from step 10}} Output: Append to CHANGELOG.md

### STEP 12: JANITOR (Cleanup)

Output: temp files what was tidied

### STEP 13: LEARNING (Record)

Input: All outputs from pipeline Output: what we learned how to prevent similar issues

### STEP 14: WARDEN (Final Check)

Input: All changes made Output: true/false any scope creep that snuck in

## EXECUTION COMMAND

When `/pipeline` is invoked with arguments, execute ALL steps above in sequence.
Pass outputs between steps using the XML tags shown.
STOP on any blocking failure and report to user.

## THE FULL CHAIN

### PHASE 1: Analysis & Planning
```yaml
step: 1
clauditos: [Orchestrator, Warden, Architect, Purpose]
flow:
  1. Orchestrator: Receive task, coordinate team
  2. Warden: Check for scope creep, enforce standards
  3. Architect: Design technical approach
  4. Purpose: Verify alignment with mission
hooks:
  - TASK_RECEIVED → Orchestrator
  - SCOPE_CHECKED → Warden
  - APPROACH_DESIGNED → Architect
  - MISSION_ALIGNED → Purpose

PHASE 2: Fix Cycle (ITERATIVE)

step: 2
clauditos: [Fixer, Debugger, Validator, Critic]
flow:
  1. Fixer: Apply minimal fix
  2. Debugger: Test the fix works
  3. Validator: Quality gate check
  4. Critic: Find weaknesses

  LOOP until Critic approves:
    → Fixer (address weaknesses)
    → Debugger (retest)
    → Validator (revalidate)
    → Critic (review again)

max_iterations: 3
hooks:
  - FIX_APPLIED → Fixer
  - TESTS_PASSED → Debugger
  - QUALITY_GATE_PASSED → Validator
  - CRITIC_APPROVED → Critic (or CRITIC_REJECTED with constraints)

PHASE 3: Deep Verification

step: 3
clauditos: [CI/CD, Telemetry, Validator, Forensics]
flow:
  1. CI/CD: Run full test suite
  2. Telemetry: Check metrics, no regressions
  3. Validator: Final quality check
  4. Forensics: Hunt for landmines introduced

  IF Forensics finds issues:
    → Fixer → Debugger → Validator → Critic → Forensics
    (mini fix cycle)

hooks:
  -