@uluops/setup 0.2.0 → 0.4.0

package/assets/commands/workflows/prompt-audit.md

@@ -1,754 +1,44 @@
----
-name: prompt-audit
-description: Comprehensive audit of agent definitions, commands, or workflows. Uses prompt-pattern-analyzer for ecosystem context, prompt-engineer for validation, and prompt-quality-validator for best practices analysis. Use when reviewing existing prompts or before significant changes.
-tools: Read, Grep, Glob, Bash
-model: sonnet
----
-
-# Prompt Audit
-
-Comprehensive audit of agent definitions, commands, or workflows. Uses prompt-pattern-analyzer for ecosystem context, prompt-engineer for validation, and prompt-quality-validator for best practices analysis. Use when reviewing existing prompts or before significant changes.
-
-
-**Philosophy**: Context-aware auditing. Understand ecosystem conventions first, then validate and improve with that context.
-### Prompt Audit vs prompt-validate
-
-| Aspect | Prompt Audit | prompt-validate |
-|--------|------|------|
-| Focus | Full audit with ecosystem context | Quick validation only |
-| Duration | 6-15 minutes | 2-5 minutes |
-| Agents Used | pattern-analyzer + prompt-engineer + prompt-quality | prompt-engineer only |
-| Output | Features list + conditional refactored draft | Pass/fail with suggestions |
-
----
-
-## Workflow Overview
-
-```
-┌──────────────────┐
-│ pattern-analyzer  │ ◄── Ecosystem context
-└────────┬─────────┘
-         │
-┌────────▼─────────┐
-│ prompt-engineer   │ ◄── Validate with context (gate)
-│ + type-specific   │     Includes inline structural checks
-│   checks          │     based on artifact type
-└────────┬─────────┘
-         │
-┌────────▼─────────┐
-│ prompt-quality    │ ◄── Best practices review
-│   validator       │
-└────────┬─────────┘
-         │
-═════════▼══════════
- PERSIST TO TRACKER
- + Conditional draft
- (if score < 75)
-════════════════════
-
-```
-
-Phase 0 establishes ecosystem conventions that inform all subsequent phases
-Type-specific structural checks are performed inline by the orchestrator between prompt-engineer and prompt-quality phases
-Refactored draft is a conditional output artifact generated only if prompt-engineer score < 75
-
-Duration: 6-15 minutes (includes ecosystem analysis)
-### Token Estimation
-
-| Scope | Input Tokens | Output Tokens |
-|-------|-------------|---------------|
-| Single agent prompt | ~35,000 | ~4,000 |
-| Workflow command | ~45,000 | ~6,000 |
-| Large workflow (500+ lines) | ~60,000 | ~8,000 |
-**Cost Drivers**:
-- Ecosystem size: more agents = more pattern analysis tokens
-- Prompt length: longer prompts = more validation tokens
-- Issue count: more findings = more detailed output
-Prompt audit is always sequential—each phase depends on prior context.
-Phase 0 ecosystem context informs Phase 1 scoring benchmarks.
-Phase 2 type checks depend on Phase 1 results.
-
----
-
-## Agent Handoff Formats
-
-Each agent passes structured data to the next in the pipeline:
-
-| From | To | Passes | Expects |
-|------|-----|--------|---------|
-| Pattern Analyzer | Prompt Engineer | Ecosystem conventions, threshold benchmarks, decision pair standards, expected sections | Context-aware validation scoring using ecosystem baselines |
-| Prompt Engineer | Prompt Quality Validator | Score, category breakdown, identified issues, structural compliance results | Best practice improvements with specific rewrite suggestions |
-
-**Handoff Contract:**
-- Pattern Analyzer provides context that informs all subsequent scoring
-- Each phase builds on prior findings rather than re-checking
-- Critical failures from Prompt Engineer block the pipeline
-- All findings feed into tracker persistence regardless of outcome
-
----
-
-## Pre-Flight: Target Detection and Configuration
-
-Before running agents, determine the target path and which optional validators should run.
-
-### Context Detection
-
-**Detection criteria**: A detector returns TRUE if its command exits with code 0.
-
-| Detector ID | Description |
-|-------------|-------------|
-| `is_agent_definition` | Run command: echo "{{ file }}" | grep -qE "agents/.*-agent\.md$" |
-| `is_agent_command` | Run command: echo "{{ file }}" | grep -qE "commands/agents/" |
-| `is_workflow_command` | Run command: echo "{{ file }}" | grep -qE "commands/workflows/" |
-| `is_general_command` | Run command: echo "{{ file }}" | grep -qE "commands/" | grep -vE "commands/(agents|workflows)/" |
-
-**is_agent_definition**:
-```bash
-echo "{{ file }}" | grep -qE "agents/.*-agent\.md$" && echo "DETECTED" || echo "NOT DETECTED"
-```
-
-**is_agent_command**:
-```bash
-echo "{{ file }}" | grep -qE "commands/agents/" && echo "DETECTED" || echo "NOT DETECTED"
-```
-
-**is_workflow_command**:
-```bash
-echo "{{ file }}" | grep -qE "commands/workflows/" && echo "DETECTED" || echo "NOT DETECTED"
-```
-
-**is_general_command**:
-```bash
-echo "{{ file }}" | grep -qE "commands/" | grep -vE "commands/(agents|workflows)/" && echo "DETECTED" || echo "NOT DETECTED"
-```
-
-### Metadata Extraction
-
-Before running phases, extract the following metadata from the target:
-
-| Step | Type | Description |
-|------|------|-------------|
-| `frontmatter_metadata` | frontmatter | Extract name, description, model, and tools from target file frontmatter |
-| `section_count` | command | Count ## headings in the target file |
-| `file_size` | command | Count lines in the target file |
-
-**frontmatter_metadata**: Extract name, description, model, tools from `{{ file }}`
-**section_count**:
-```bash
-grep -c '^## ' {{ file }}
-```
-
-**file_size**:
-```bash
-wc -l < {{ file }}
-```
-
-
----
-
-## Arguments
-
-### Positional Arguments
-
-| Argument | Required | Description |
-|----------|----------|-------------|
-| file | Yes | Path to the prompt artifact to audit (agent, command, or workflow file) |
-
-
-### Usage Examples
-
-| Command | Behavior |
-|---------|----------|
-| `/workflows:prompt-audit agents/security-analyst-agent.md` | Audits agent definition with ecosystem context |
-| `/workflows:prompt-audit commands/agents/validate.md` | Audits agent invocation command |
-| `/workflows:prompt-audit commands/workflows/ship.md` | Audits workflow command |
-
----
-
-## Execution Mode Selection
-
-
-| Mode | Description | Best For |
-|------|-------------|----------|
-| Sequential | - | - |
-| Parallel | - | - |
-
-
-Prompt audit is always sequential—each phase depends on prior context.
-Phase 0 ecosystem context informs Phase 1 scoring benchmarks.
-Phase 2 type checks depend on Phase 1 results.
-
----
-
-## Execution
-
-Run each agent in sequence (or parallel groups if selected). Stop and fix if any agent fails. **Collect all recommendations for tracker persistence.**
-
-### Phase 1: Ecosystem Pattern Analysis
-**Commands**: pattern-analyzer@1.0.0
-
-**Invoke via Task tool:**
-```
-Task(
-  subagent_type: "pattern-analyzer",
-  prompt: "[validator:pattern-analyzer] Validate {TARGET_DIRECTORY}. Return structured JSON OUTPUT.",
-  description: "Pattern Analyzer"
-)
-```
-
-**Gate**: threshold >= 50, on fail: warn
-
-**Focus**:
-- Discover all prompt artifacts (agents, commands, workflows)
-- Extract scoring patterns and point allocations
-- Identify decision vocabulary conventions
-- Analyze threshold standards by agent type
-- Detect structural patterns and inconsistencies
-
-**Capture for tracker**: Ecosystem context (conventions, thresholds, patterns) for downstream phases.
-
-**If failing**: Ecosystem data is limited (<5 agents). Proceed with generic conventions.
-
-**Decision criteria**:
-- INSIGHTFUL (✅): Score ≥75 AND ≥50% ecosystem coverage
-- INCOMPLETE (⚠️): Score <75 OR <5 agents in ecosystem
-
-**Alternatives**:
-- `strategy-analyst@1.0.0` — Must override model to sonnet — default is opus. Use when deeper strategic analysis is needed.
-
-**Key Outputs**:
-
-| Output | Downstream Usage |
-|--------|-----------------|
-| threshold_convention | Benchmark for Phase 1 scoring — e.g., expected >=75 for validators |
-| decision_pair_convention | Consistency check in Phase 1 — e.g., PASS/FAIL, DEPLOY/REVISE pairs |
-| expected_sections | Type-specific validation — e.g., Mission, Output Format, Decision for agents |
-| common_failure_modes | Focus areas for Phase 2 improvement suggestions |
-
----
-
-### Phase 2: Prompt Engineer Validation
-**Commands**: prompt-validate@1.0.0
-
-**Invoke via Task tool:**
-```
-Task(
-  subagent_type: "prompt-validate",
-  prompt: "[validator:prompt-validate] Validate {TARGET_DIRECTORY}. Return structured JSON OUTPUT.",
-  description: "Prompt Validator"
-)
-```
-
-**Gate**: threshold >= 75, on fail: stop
-
-**Why this threshold?** Prompts below 75 produce inconsistent agent behavior. The threshold ensures minimum deployment quality.
-
-**Focus**:
-- Clarity & Specificity (25 pts)
-- Structure & Organization (20 pts)
-- Completeness (25 pts)
-- Effectiveness (20 pts)
-- Consistency (10 pts)
-- Type-specific structural compliance (inline checks per artifact type)
-
-**Capture for tracker**: Overall score, category breakdown, vague language instances with line numbers, missing sections, improvement suggestions, and type-specific structural compliance.
-
-**If failing**: Apply critical fixes from the improvement suggestions. Prompt does not meet deployment quality.
-
-**Decision criteria**:
-- DEPLOY (✅): Score ≥75 AND no critical issues AND structural checks pass
-- REVISE (❌): Score <75 OR critical issues present OR structural elements missing
-
-### For Agent Definitions
-
-**Runs when**: `is_agent_definition`
-
-Check for required elements:
-
-| Check | What to Look For | Report |
-|-------|------------------|--------|
-| Required sections | "Mission", "Output Format", "Decision" | [OK]/[X] for each |
-| Scoring framework | Contains "points", "pts", or "/100" | Show first 5 matches |
-| Auto-fail conditions | Contains "auto.*fail", "critical.*issue", "must.*fix" | List if found |
-| Quality gate | Contains checklist or gate criteria | List if found |
-| Edge case handling | Has subsections (### headings) | Count found |
-
-### For Agent Commands
-
-**Runs when**: `is_agent_command`
-
-Check for required elements:
-
-| Check | What to Look For | Report |
-|-------|------------------|--------|
-| Agent reference | Contains "agents:" or "agent-" | [OK]/[X] |
-| Argument handling | Contains "ARGUMENTS", "<directory>", "<path>" | [OK]/[X] |
-| Threshold mention | Contains ">=[0-9]+", "Threshold", or "Score" | [OK]/[X] |
-
-### For Workflows
-
-**Runs when**: `is_workflow_command`
-
-Check for required elements:
-
-| Check | What to Look For | Report |
-|-------|------------------|--------|
-| Phase structure | Contains "### Phase" or "Phase [0-9]" | List phases found |
-| Agent invocations | Contains "/agents:" or "agent" references | List agents used |
-| Decision gates | Contains "PASS", "FAIL", "STOP", "Continue", "proceed" | [OK]/[X] |
-| Summary format | Contains "Summary", "Report", or "Result" | [OK]/[X] |
-
-**Depends on**: pattern-analyzer
-
----
-
-### Phase 3: Prompt Quality Best Practices
-**Commands**: prompt-quality@1.0.0
-
-**Invoke via Task tool:**
-```
-Task(
-  subagent_type: "prompt-quality",
-  prompt: "[validator:prompt-quality] Validate {TARGET_DIRECTORY}. Return structured JSON OUTPUT.",
-  description: "Prompt Quality Validator"
-)
-```
-
-**Gate**: threshold >= 70, on fail: warn
-
-**Focus**:
-- Clarity: Vague language → specific alternatives
-- Structure: Missing sections → template additions
-- Completeness: Missing edge cases → defined behaviors
-- Effectiveness: Subjective criteria → objective measures
-
-**Capture for tracker**: Best practices analysis with specific rewrite suggestions.
-
-**If failing**: Review improvement suggestions. Prompt is functional but could be more effective.
-
-**Decision criteria**:
-- STRONG (✅): Score ≥75 AND no high-priority findings
-- ADEQUATE (⚠️): Score 60-74
-- WEAK (❌): Score <60
-
-**Depends on**: prompt-engineer
-
----
-
-
----
-
-## Summary Report
-
-After all phases complete, summarize:
-
-```
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-SUMMARY
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-Target: [path]
-Run: [N]
-
-┌─────────────────────┬────────┬────────────┐
-│ Agent               │ Score  │ Status     │
-├─────────────────────┼────────┼────────────┤
-│ ...                 │ [X]/100│ ✅/❌/⏭️   │
-└─────────────────────┴────────┴────────────┘
-
-```
-
-
-### Scoring
-
-**Method**: weighted_average
-
-| Phase | Weight |
-|-------|--------|
-| pattern-analyzer | 15% |
-| prompt-engineer | 55% |
-| prompt-quality | 30% |
-
-
-## Output: Consolidated Report
-
-```
-+=============================================================================+
-|                          PROMPT AUDIT REPORT                                |
-+=============================================================================+
+---
+name: prompt-audit
+description: Comprehensive prompt audit with ecosystem context. Pattern analyzer provides ecosystem baseline, prompt engineer validates structure and scoring, prompt quality validates clarity and effectiveness. Sequential with gating.
+tools: Read, Grep, Glob, Bash
+model: opus
+---
 
-File: $ARGUMENTS
-Type: {agent | command | workflow}
-Name: {extracted name}
+# Prompt Audit
 
-==============================================================================
+> Comprehensive prompt audit with ecosystem context. Pattern analyzer provides ecosystem baseline, prompt engineer validates structure and scoring, prompt quality validates clarity and effectiveness. Sequential with gating.
 
-PHASE 0: ECOSYSTEM PATTERN ANALYSIS
+Duration: 6-15 minutes
+**Arguments**: `file` (required)
+## Pre-Flight Detection
 
-Ecosystem Inventory:
-- Agents: [N]
-- Commands: [N]
-- Workflows: [N]
+- **is_agent_definition**: `grep -rqE --include='{{ file }}'  '^(name|description|type):' . 2>/dev/null`
+- **is_agent_command**: `grep -rqE --include='{{ file }}'  'subagent_type|\[validator:' . 2>/dev/null`
+## Execution
 
-Ecosystem Conventions:
-- Threshold standard: >=[X] ([N]% of agents)
-- Decision pair: [KEYWORD_A]/[KEYWORD_B]
-- Required sections: [list]
-- Point distribution: [pattern]
-
-Consistency Check:
-{If target diverges from conventions:}
-- Target uses [X] but ecosystem standard is [Y]
-- Consider aligning for consistency
-
-{If target follows conventions:}
-- [OK] Target aligns with ecosystem conventions
-
-Status: [OK INSIGHTFUL | WARNING INCOMPLETE]
-
--------------------------------------------------------------------------------
-
-PHASE 1: PROMPT ENGINEER VALIDATION
-
-Score: [X]/100
-
-Clarity & Specificity:  [X]/25
-Structure:              [X]/20
-Completeness:           [X]/25
-Effectiveness:          [X]/20
-Consistency:            [X]/10
-
-Status: [DEPLOY | REVISE]
-
--------------------------------------------------------------------------------
-
-PHASE 2: TYPE-SPECIFIC CHECKS
-
-Type: {artifact_type}
-
-Required Elements:
-[OK] {present element}
-[OK] {present element}
-[X] {missing element}
-
-Convention Compliance:
-| Element | Found | Expected | Status |
-|---------|-------|----------|--------|
-| Decision keywords | X | Y | OK/X |
-| Threshold format | X | Y | OK/X |
-| Scoring format | X | Y | OK/X |
-
--------------------------------------------------------------------------------
-
-PHASE 3: IMPROVEMENT ANALYSIS
-
-CRITICAL (Must Fix):
-1. [Issue]: Line [N]
-   Problem: {why it matters}
-   Fix: {specific rewrite}
-
-RECOMMENDED (Should Fix):
-1. [Issue]: Line [N]
-   Problem: {impact}
-   Fix: {suggestion}
-
-SUGGESTIONS (Consider):
-1. [Enhancement]: {description}
-
--------------------------------------------------------------------------------
-
-IMPROVEMENT EXAMPLES
-
-### Example 1: {issue type}
-
-**Location:** Line [N]
-
-**Current:**
-{current text}
-
-**Improved:**
-{better text}
-
-**Why better:** {explanation}
-
--------------------------------------------------------------------------------
-
-PHASE 4: REFACTORED DRAFT
-
-{If score < 75, include full refactored prompt}
-{If score >= 75, show "Not required - prompt passes validation"}
-
-===============================================================================
-
-AUDIT DECISION
-
-[APPROVED - Ready for use]
-  No changes required. Prompt meets project standards.
-
-OR
-
-[IMPROVEMENTS AVAILABLE - Optional enhancements]
-  Prompt is functional but could be improved.
-  See SUGGESTIONS section above.
-
-OR
-
-[REVISION REQUIRED - Must fix before deployment]
-  Critical issues found. See CRITICAL section.
-  Refactored draft provided in Phase 4.
-
-===============================================================================
-
-NEXT STEPS
-
-{If APPROVED}
-- No action required
-- Consider suggestions for future iterations
-
-{If IMPROVEMENTS AVAILABLE}
-- Review suggestions
-- Apply changes that add value
-- Re-run audit to verify (optional)
-
-{If REVISION REQUIRED}
-1. Apply critical fixes from Phase 3
-2. OR use refactored draft from Phase 4
-3. Re-run: /workflows:prompt-audit $ARGUMENTS
-4. Iterate until APPROVED
-
-+=============================================================================+
-
-```
-
-### Decision Criteria
-
-| Score | Issues | Decision |
-|-------|--------|----------|
-| >=85 | None critical | [OK] APPROVED |
-| 75-84 | None critical | [FIX] IMPROVEMENTS AVAILABLE |
-| <75 OR critical issues | Any | [LOOP] REVISION REQUIRED |
-
-**[LOOP] REVISION REQUIRED Auto-Triggers:**
-- Undefined or vague mission statement
-- No output format specification
-- Conflicting instructions
-- Missing scoring/threshold for validation agents
-- Decision criteria are purely subjective
-
-
----
-
-## Final Phase: Outputs (MANDATORY)
-
-**This phase runs regardless of pass/fail status.** All agent recommendations must be captured.
-
-### Artifacts
-
-**features-list** (markdown):
-Consolidated recommendations from all phases with scoring breakdown, category analysis, and actionable improvement items for tracker persistence.
-
-Generate a timestamp for the filename:
-
-```bash
-TIMESTAMP=$(date +%Y-%m-%dT%H-%M-%S)
-echo "Timestamp: $TIMESTAMP"
 ```
-
-Write file to: `docs/{{ features_file }}`
-
-**Content template:**
-
-# Prompt Audit: {{ file_name }}
-
-**Date**: {{ timestamp }}
-**Target**: {{ file_path }}
-**Type**: {{ artifact_type }}
-
-## Scores
-
-| Phase | Score | Status |
-|-------|-------|--------|
-| Pattern Analyzer | {score}/100 | {PASS/WARN} |
-| Prompt Engineer | {score}/100 | {PASS/FAIL} |
-| Prompt Quality | {score}/100 | {PASS/WARN} |
-| **Weighted Average** | **{score}/100** | |
-
-## Recommendations
-
-### Critical (fix before deployment)
-- [ ] {recommendation with file:line reference}
-
-### Suggested (review before deployment)
-- [ ] {recommendation with file:line reference}
-
-### Backlog (post-deployment)
-- [ ] {recommendation}
-
-
-**refactored-draft** (markdown):
-A refactored version of the target prompt with all critical fixes applied from Phases 0-2. Generated only when prompt-engineer score is below deployment threshold.
-**Condition**: Generated only when `phases.prompt-engineer.score < 75`
-
-Generate a timestamp for the filename:
-
-```bash
-TIMESTAMP=$(date +%Y-%m-%dT%H-%M-%S)
-echo "Timestamp: $TIMESTAMP"
+Group 1 (sequential): pattern-analyzer
+Group 2 (sequential): prompt-engineer
+Group 3 (sequential): prompt-quality
 ```
-
-Write file to: `docs/{{ file_name }}-refactored.md`
-
-#### Granularity Rules
-
-**DO NOT consolidate multiple findings into single entries.** The tracker handles deduplication via fingerprinting. Save ALL raw findings:
-
-| Agent Reports | You Save |
-|---------------|----------|
-| "N+1 pattern in 14 locations" | 14 separate recommendations, one per file:line |
-| "Missing null checks on 8 .find() calls" | 8 separate recommendations |
-| "6 endpoints missing from docs" | 6 separate recommendations |
-
-**Rule:** If validator output has a file:line reference, it becomes a separate recommendation entry.
-
-### Save to Tracker (MANDATORY — Always Save First)
-
-**CRITICAL: Always save to the tracker. Never skip this step.**
-
-After writing the markdown file, save results via `mcp__uluops-tracker__save_features_list`:
-
-```
-mcp__uluops-tracker__save_features_list({
-  project: claude-agent-workflows,
-  workflow_type: "prompt-audit",
-  timestamp: {ISO8601 timestamp},
-  validators: [
-    {
-      "name": "{Agent Name}",
-      "score": {numeric},
-      "status": "{PASS|FAIL|SKIP}",
-      "model": "{haiku|sonnet|opus}"
-    }
-    // ... one entry per agent
-  ],
-  recommendations: [
-    {
-      "validator": "{agent-name}",
-      "title": "{short title}",
-      "priority": "{critical|suggested|backlog}",
-      "severity": "{critical|high|medium|low|info}",
-      "description": "{details}",
-      "file_path": "{path}",
-      "line_number": {number}
-    }
-    // ... one entry per recommendation
-  ]
-})
-```
-
-**Priority mapping:**
-- Blocking (fix before ship) → `"critical"`
-- Warnings (review before ship) → `"suggested"`
-- Post-ship/backlog items → `"backlog"`
-
-### Post-Save Verification
-
-After saving, verify the data was persisted correctly. **These are warnings, not blockers.**
-
-**Query validation-tracker and compare to tracker_payload_length**
-
-- On mismatch: **warn** (do NOT block or re-attempt)
-- Saved count differs from payload - some recommendations may not have persisted.
-
-**Verification procedure:**
-
-1. Query the tracker for the saved run
-2. Compare saved recommendation count against your payload count
-3. If counts differ, log the discrepancy as a note — cross-phase deduplication is expected when multiple validators flag the same issue
-4. **Proceed regardless** — the save already succeeded
-
-
----
-
-
-
-
-
----
-
-## Iteration Pattern
-
-```
-Existing Prompt
-     │
-     ▼
-/workflows:prompt-audit
-     │
-     ├── APPROVED ──────▶ Done (optional improvements noted)
-     │
-     ├── IMPROVEMENTS ──▶ Apply suggestions ──▶ (optional re-audit)
-     │
-     └── REVISION ──────▶ Apply fixes OR use refactored draft
-                              │
-                              └──▶ /workflows:prompt-audit (repeat)
-
-```
-
-**Typical iterations**:
-- Well-written prompt: 1 run (APPROVED or IMPROVEMENTS)
-- Prompt with issues: 2-3 runs
-- Legacy or undocumented prompt: 2-4 runs
-
-**Report behavior across iterations**:
-- Each run creates a features list documenting current state and recommendations
-- Previous runs are preserved for audit trail
-- Resolved items from previous runs will not appear if agents no longer flag them
-- Tracker detects regressions when previously fixed issues reappear
-
----
-
-## Quick Reference
-
-| Agent | Threshold | Group |
-|-------|-----------|-------|
-| Ecosystem Pattern Analysis | threshold >= 50, on fail: warn | 1 |
-| Prompt Engineer Validation | threshold >= 75, on fail: stop | 2 |
-| Prompt Quality Best Practices | threshold >= 70, on fail: warn | 3 |
-
-
----
-
-## Troubleshooting
-
-### Phase 3 suggestions conflict with each other
-
-The prompt-quality-validator generates suggestions independently. If suggestions conflict:
-1. Prioritize CRITICAL over RECOMMENDED over SUGGESTIONS
-2. When two suggestions target the same section, choose the more specific one
-3. Use your judgment on which improves clarity most
-
-
-### Refactored draft is too different from original
-
-The refactored draft prioritizes correctness over minimal changes. If you prefer incremental updates:
-1. Apply only the CRITICAL fixes manually
-2. Re-audit to verify those fixes
-3. Then consider RECOMMENDED items one by one
-
-
-### Type detection is wrong
-
-Override by specifying in your prompt:
-"Treat this as an agent definition and audit accordingly."
-/workflows:prompt-audit agents/my-prompt.md
-
-
-### Score seems too low for a working prompt
-
-The audit is strict because prompts compound errors. A 65-score prompt may work but produces inconsistent results. The refactored draft shows what a higher-scoring version looks like.
-
-
-### Pattern analyzer shows INCOMPLETE
-
-This happens when the ecosystem has fewer than 5 agents. The audit proceeds with generic conventions instead of project-specific patterns. Results are still valid but less context-aware.
-
-
+## Phases
+
+| # | Agent | Threshold | Gate | Condition |
+|---|-------|-----------|------|-----------|
+| 1 | pattern-analyzer@latest | threshold >= 50, on fail: warn | stop | — |
+| 2 | prompt-validate@latest | threshold >= 75, on fail: stop | stop | — |
+| 3 | prompt-quality@latest | threshold >= 70, on fail: warn | stop | — |
+
+**Ecosystem Pattern Analysis**: Ecosystem scoring patterns and conventions; Decision vocabulary and threshold standards
+**Prompt Engineering Validation** (after pattern-analyzer): Clarity, structure, and completeness; Scoring framework and effectiveness
+**Prompt Quality Validation** (after prompt-engineer): Clarity and specificity; Objective effectiveness criteria
+## Scoring
+
+**Method**: weighted_average
+ — pattern-analyzer: 15%, prompt-engineer: 55%, prompt-quality: 30%
+## Results Submission
+
+Write markdown report to: `{{ target_path }}/{{ report_file }}`
+Save ALL findings to tracker via `mcp__uluops-tracker__save_features_list` with project=`{{ target_name }}`, workflow_type=`prompt-audit`, definition_type=`workflow`, definition_name=`prompt-audit`, definition_version=`2.0.0`. Include validators array (name, score, status, model) and recommendations array (validator, title, priority, severity, description, file_path, line_number). Each file:line reference becomes a separate recommendation. Priority: blocking=critical, warnings=suggested, post-ship=backlog.
+After saving, query tracker and compare counts. Mismatches from cross-phase deduplication are expected — warn only, do not re-attempt.