@hailer/mcp 1.1.12 → 1.1.13

package/.opencode/commands/integration-squad.md

@@ -1,82 +0,0 @@
----
-description: Build backend integrations with Kenji, Ivan, and Tanya
-argument-hint: "integration description"
-allowed-tools: Task, Bash, Read
----
-# Integration Squad
-
-Sequential pipeline: Schema discovery, build integration, then verify with tests.
-
-**Agents:**
-1. **Kenji** - Schema discovery (workflow fields, types, IDs)
-2. **Ivan** - Build integration (monolith webhooks, scheduled jobs, automations)
-3. **Tanya** - Test runner (verify integration with tests)
-
-**Goal:** $ARGUMENTS
-
-## Protocol
-
-### Step 1: Kenji (Schema Discovery)
-
-Start with Kenji to gather workflow schema and field IDs needed for the integration:
-
-```
-Task(subagent_type="agent-kenji-data-reader", prompt="List all workflows and their fields relevant to: $ARGUMENTS. Include field IDs, types, and any existing integrations. Also identify webhook URLs and activity phase names needed.")
-```
-
-Wait for result. Save the schema output - Ivan needs these IDs.
-
-### Step 2: Ivan (Build Integration)
-
-Spawn Ivan with the schema context from Kenji:
-
-```
-Task(subagent_type="agent-ivan-monolith", prompt="Build a backend integration for: $ARGUMENTS.\n\nWorkflow schema from Kenji:\n[PASTE KENJI'S OUTPUT - workflow IDs, field IDs, types, phases]\n\nImplement in the hailer-integration monolith. Use correct IDs for webhooks, scheduled jobs, or automations.")
-```
-
-Wait for result. If `"status": "ready_to_push"`:
-- Run the push commands Ivan provides
-
-### Step 3: Tanya (Test Runner)
-
-Spawn Tanya to verify the integration works:
-
-```
-Task(subagent_type="agent-tanya-test-runner", prompt="Test the integration for: $ARGUMENTS.\n\nTest coverage needed:\n- Webhook payload validation (if webhook)\n- Scheduled job execution (if scheduled)\n- Activity creation/phase transitions\n- Error handling and edge cases\n\nRun tests and report pass/fail counts.")
-```
-
-Wait for result. If tests fail, report which ones and offer to have Ivan fix issues.
-
-### Step 4: Report
-
-```markdown
-## Integration Squad Complete
-
-### Kenji (Schema Discovery)
-- Workflows found: [list]
-- Fields mapped: [count]
-- Phases identified: [list]
-- Existing integrations: [list or "none"]
-
-### Ivan (Monolith Integration)
-- Integration type: [webhook/scheduled/automation]
-- Files created/modified: [list]
-- Endpoints: [list if webhook]
-- Commands run: [list]
-
-### Tanya (Tests)
-- Tests: X passed, X failed
-- Build: Pass/Fail
-- [Failed test details if any]
-
-### Next Steps
-- Deploy to staging environment
-- Test end-to-end in Hailer
-- Monitor for errors in production logs
-```
-
-## Notes
-
-- Kenji must run first to get real IDs - never let Ivan guess IDs
-- Each agent in the pipeline depends on output from the previous step
-- If tests fail, Ivan may need to revise the integration code

package/.opencode/commands/janitor-squad.md

@@ -1,167 +0,0 @@
----
-description: Deep project cleanup - code, config, and data patterns
-argument-hint: [scope or blank for full project]
-allowed-tools: Task, Bash
----
-# Janitor Squad
-
-Deep project cleanup across code, configuration, and data. Discovers dead patterns and auto-fixes clear-cut removals.
-
-**Agents:**
-- **Lars** - Dead code detection (unused imports, unreachable code, type errors via LSP)
-- **Bjorn** - Configuration audit (deprecated fields, broken references, unused workflows)
-- **Viktor** - Data analysis (unpopulated fields, zero-activity workflows, silent insights)
-- **Simple Writer** - Auto-fix confirmed dead code and unused imports (fix loop)
-- **Code Simplifier** - Simplify complex code blocks identified during discovery (fix loop)
-
-**Target:** $ARGUMENTS (if blank, scan full project)
-
-**Changed files:** Orchestrator runs `git diff --name-only HEAD` to scope if needed.
-
-## Protocol
-
-### Step 1: Determine Scope
-
-If `$ARGUMENTS` has specific files/directories, use those as the target.
-Otherwise, scan the entire project.
-
-If `$ARGUMENTS` contains `--code-only`:
-- Skip Bjorn and Viktor
-- Only run Lars in discovery phase
-
-If `$ARGUMENTS` contains `--config-only`:
-- Skip Lars and Viktor
-- Only run Bjorn in discovery phase
-
-If `$ARGUMENTS` contains `--report-only`:
-- Skip all fix loop iterations
-- Report findings only (no auto-fixes)
-
-### Step 2: Launch Discovery Agents in Parallel
-
-**Set:** `iteration = 1`
-
-Spawn all three discovery agents simultaneously using multiple Task tool calls in a single message:
-
-**Lars:**
-```
-Task(subagent_type="agent-lars-code-inspector", prompt="Inspect the project for dead code patterns: [TARGET]. Find:\n- Unused imports\n- Unreachable code blocks\n- Unused variables and functions\n- Type errors\n\nUse LSP if available, fall back to tsc/eslint. Return detailed findings grouped by file.")
-```
-
-**Bjorn:**
-```
-Task(subagent_type="agent-bjorn-config-audit", prompt="Audit workspace configuration for zombie patterns: [TARGET]. Check:\n- Deprecated fields in workspace/fields.ts\n- Broken field references in phases\n- Unused workflows (defined but zero activities)\n- Broken enum references\n- Orphaned configuration entries\n\nReturn issues grouped by category with severity.")
-```
-
-**Viktor:**
-```
-Task(subagent_type="agent-viktor-sql-insights", prompt="Analyze data patterns in workflows: [TARGET]. Find:\n- Fields that are defined but never populated (zero values across all activities)\n- Workflows with zero activities (created but unused)\n- Insights that return empty data (broken or obsolete)\n- Calculated fields that never execute\n- Orphaned data patterns\n\nReturn findings with field IDs, workflow IDs, and impact estimates.")
-```
-
-### Step 3: Categorize Findings
-
-Collect results from all three agents. Categorize each finding as:
-
-**AUTO-FIXABLE** (Simple Writer can safely remove):
-- Unused imports identified by Lars
-- Dead code blocks with no side effects
-- Workflow definitions with zero activities (from Viktor + Bjorn)
-- Deprecated enum references (from Bjorn)
-- Never-populated fields that can be safely archived (from Viktor + Bjorn)
-
-**NEEDS-HUMAN-REVIEW** (escalate to user):
-- Architecture concerns
-- Fields that are rarely used but intentional
-- Complex logic entangled with dead code
-- Data cleanup that affects reporting history
-- Performance optimizations requiring design changes
-
-### Step 4: Auto-Fix Loop (if auto-fixable items found AND `--report-only` not set AND `iteration < 3`)
-
-If critical auto-fixable items exist, spawn two agents to clean up. If Simple Writer's fixes introduce new issues detected by re-audit, escalate remaining issues to user after max 2 iterations.
-
-**Simple Writer:**
-```
-Task(subagent_type="agent-simple-writer", prompt="Remove confirmed dead code and unused imports from the project:\n\n[LIST AUTO-FIXABLE ITEMS WITH FILE PATHS AND LINE NUMBERS]\n\nFrom Lars:\n- [Unused imports with files]\n- [Dead code blocks with files]\n\nFrom Bjorn:\n- [Deprecated enum references with files]\n- [Unused workflow definitions with IDs]\n\nFrom Viktor:\n- [Never-populated field definitions with IDs]\n\nApply minimal removals only. Do not refactor or improve beyond confirmed dead patterns.")
-```
-
-**Code Simplifier:**
-```
-Task(subagent_type="agent-code-simplifier", prompt="After dead code removal, simplify complex code blocks identified as difficult to maintain: [TARGET].\n\nFocus on:\n- Variable naming clarity\n- Function complexity reduction\n- Comment clarity around remaining logic\n- Code organization in affected files\n\nReturn refactoring suggestions with before/after examples.")
-```
-
-After both finish:
-- Increment `iteration`
-- Go back to **Step 2** to re-run all discovery agents
-- On re-discovery, tell agents to verify previous issues are resolved
-- Max 2 re-check iterations (3 total discovery runs)
-
-### Step 5: Report
-
-```markdown
-## Janitor Squad Report
-
-### Loop Summary
-- Discovery iterations: [count]
-- Auto-fixes applied: [yes/no]
-- Items removed: [count by category]
-
-### Lars (Dead Code Detection)
-- Unused imports: X (Y auto-removed)
-- Dead code blocks: X (Y auto-removed)
-- Type errors: X
-- [Remaining issues grouped by file]
-
-### Bjorn (Config Audit)
-- Deprecated fields: X
-- Broken references: X (Y auto-removed)
-- Unused workflows: X (Y auto-removed)
-- Orphaned config: X
-- [Remaining issues grouped by category]
-
-### Viktor (Data Analysis)
-- Never-populated fields: X (Y marked for archival)
-- Zero-activity workflows: X (Y removed)
-- Silent insights: X
-- Orphaned data patterns: X
-- [Remaining items grouped by workflow]
-
-### Auto-Fixes Applied
-[If any fixes were made:]
-- [File/ID]: [what was removed/archived]
-- [File/ID]: [what was removed/archived]
-
-### Remaining Issues (Manual Action Needed)
-[Issues that require human review]
-- [Category]: [Details]
-
-### Recommended Cleanup Order
-1. [Remove these orphaned items first]
-2. [Archive these fields next]
-3. [Review these design concerns]
-
-### Summary
-[Overall health assessment]
-[Estimated time to apply remaining cleanup]
-```
-
-If any agent fails or times out, report partial results from successful agents, note which failed, and offer to re-run individually.
-
-## Options
-
-| Flag | Effect |
-|------|--------|
-| `--code-only` | Skip Bjorn and Viktor, just code cleanup via Lars |
-| `--config-only` | Skip Lars and Viktor, just workspace config audit via Bjorn |
-| `--report-only` | Report findings only, no auto-fix loop (discovery phase only) |
-| `--bg` | Run all discovery agents in background mode |
-
-If `--bg` is present, launch all discovery agents with `run_in_background: true` and tell the user they'll be notified when discovery is complete, then proceed with fix loop when they're ready.
-
-## Notes
-
-- Max 2 re-discovery iterations (3 total) to prevent infinite loops
-- Simple Writer only removes clear-cut dead code - ambiguous items go to user review
-- Data cleanup (Viktor findings) may affect reports/history - always prompt user before removal
-- Each re-discovery is a full parallel sweep across all enabled agents
-- The `--report-only` flag gives discovery results without auto-fixes

package/.opencode/commands/learn-auto.md

@@ -1,120 +0,0 @@
----
-description: Scan conversation for learnings and capture to inbox
-argument-hint: [optional focus area]
----
-# Auto-Learn
-
-Scan conversation for potential learnings, present for confirmation, write to inbox.
-
-## Trigger Patterns
-
-Scan conversation for these signals:
-
-### 1. Error → Fix Sequences
-```
-Pattern: Error/failure message followed by successful resolution
-Signal: "Error:", "failed", "doesn't work" → later "fixed", "working now", "solved"
-Extract: What was wrong, what fixed it
-```
-
-### 2. User Corrections
-```
-Pattern: User corrects agent output or approach
-Signal: "No, it should be...", "Actually, use...", "That's wrong, do..."
-Extract: Wrong approach, correct approach
-```
-
-### 3. Explicit Learning Signals
-```
-Pattern: User explicitly notes something important
-Signal: "Good to know", "I'll remember that", "Important:", "Note to self"
-Extract: The insight
-```
-
-### 4. Agent Failures
-```
-Pattern: Agent returned error or wrong result
-Signal: "status": "error", validation failures, user says "that's not right"
-Extract: What failed, why, how to prevent
-```
-
-### 5. Repeated Attempts
-```
-Pattern: Same operation tried multiple times before success
-Signal: 2+ similar tool calls, retries with different parameters
-Extract: What didn't work, what finally worked
-```
-
-## Instructions
-
-### Step 1: Scan Conversation
-
-Review the conversation for patterns above. For each potential learning, note:
-- **Type**: error-fix | correction | insight | agent-failure | retry-pattern
-- **Context**: What was happening
-- **Learning**: The key takeaway
-- **Category**: agent/X | skill/X | pattern | gotcha | bug
-
-### Step 2: Present Findings
-
-If learnings found, present to user:
-
-```
-Found [N] potential learnings from this session:
-
-1. [agent/kenji] Field lookup should check enums.ts first
-   Context: Kenji made API call when data was available locally
-
-2. [pattern] Date fields from Hailer come pre-formatted
-   Context: Tried to format already-formatted date, got wrong output
-
-3. [gotcha] Push must complete before pull
-   Context: Pull overwrote local changes
-
-Confirm which to save? (all / 1,2 / none)
-```
-
-Use AskUserQuestion:
-- "Save all"
-- "Select specific (enter numbers)"
-- "Skip - nothing worth saving"
-
-### Step 3: Write to Inbox
-
-For confirmed learnings, append to `inbox.md`:
-
-```markdown
-## Pending
-
-- [<category>] <learning description> _(from: <project>, <date>, auto-learn)_
-```
-
-If in hub repo (hailer-claude-reference), write directly to `inbox.md`.
-If in project repo, write to `docs/learnings.md` or project's inbox location.
-
-### Step 4: Report
-
-```
-Captured [N] learnings to inbox:
-- [agent/kenji] Field lookup should check enums.ts first
-- [pattern] Date fields from Hailer come pre-formatted
-
-Review later with /learn to apply fixes, or let Ada auto-process.
-```
-
-## No Learnings Found
-
-If conversation has no clear learnings:
-
-```
-No obvious learnings detected in this session.
-
-To capture something manually: /learn <category> <description>
-```
-
-## Notes
-
-- This is DETECTION only - doesn't apply fixes (that's /learn or Ada's job)
-- User confirms before anything is saved
-- Bias toward false negatives (miss some) rather than false positives (noise)
-- Works best at end of session when there's context to analyze

package/.opencode/commands/learn.md

@@ -1,120 +0,0 @@
----
-description: Capture learning and apply fix to the right place
-argument-hint: <category> <description>
----
-# Learn & Apply
-
-Capture a learning, ask clarifying questions, then apply the fix to the right place.
-
-## Usage
-`/learn <category> <description>`
-
-## Categories
-- `agent/<name>` - Agent behavior issue (e.g., `agent/kenji`)
-- `skill/<name>` - Skill missing pattern (e.g., `skill/function-fields`)
-- `pattern` - General pattern to document
-- `bug` - Bug in code/config
-- `gotcha` - Common mistake to prevent
-
-## Examples
-```
-/learn agent/kenji should check enums.ts before API calls
-/learn skill/insights JOIN syntax was wrong for linked fields
-/learn pattern always verify push succeeded before continuing
-/learn gotcha phase IDs change when workflow is copied
-```
-
-## Instructions
-
-### Step 1: Parse & Understand
-
-Parse the input:
-- Category: first word (e.g., `agent/kenji`)
-- Description: rest of the text
-
-### Step 2: Ask Clarifying Questions
-
-Use AskUserQuestion to gather details:
-
-**For agent issues:**
-- "What was the symptom?" (wrong output, error, missed step)
-- "What should the agent do differently?"
-- "Should this be a rule, a skill addition, or both?"
-
-**For skill issues:**
-- "What was missing or wrong?"
-- "Can you give an example of correct vs incorrect?"
-
-**For patterns/gotchas:**
-- "What went wrong when this wasn't followed?"
-- "Where should this be documented?" (skill, agent, CLAUDE.md)
-
-### Step 3: Determine Target
-
-Based on category and answers, decide where the fix goes:
-
-| Category | Likely Target |
-|----------|---------------|
-| `agent/X` | `.claude/agents/agent-X.md` |
-| `skill/X` | `.claude/skills/X/SKILL.md` |
-| `pattern` | Relevant skill or CLAUDE.md |
-| `bug` | Code fix + possibly skill |
-| `gotcha` | Skill or agent rules section |
-
-### Step 4: Apply the Fix
-
-**Option A: Direct edit (if simple)**
-- Edit the target file directly
-- Add to appropriate section (rules, patterns, examples)
-
-**Option B: Delegate to Ada (if complex)**
-For complex changes, invoke Ada agent:
-```
-Task(subagent_type="agent-ada-skill-builder", prompt={
-  "task": "apply_learning",
-  "category": "<category>",
-  "learning": "<description>",
-  "details": "<clarification answers>",
-  "target": "<determined target>"
-})
-```
-
-### Step 5: Log & Confirm
-
-1. Log to `docs/learnings.md` for history:
-   ```markdown
-   - [<category>] <description> → Applied to <target> _(YYYY-MM-DD)_
-   ```
-
-2. Report what was done:
-   ```
-   Learning applied:
-   - Category: agent/kenji
-   - Fix: Added rule "Check enums.ts before making API calls"
-   - Location: .claude/agents/agent-kenji-data-reader.md
-   ```
-
-## File Template
-
-If creating `docs/learnings.md`:
-
-```markdown
-# Applied Learnings
-
-History of fixes applied from learnings.
-
----
-
-## Log
-
-- [agent/kenji] Check enums.ts first → Added to rules _(2026-01-31)_
-- [skill/insights] JOIN syntax example → Added to SKILL.md _(2026-01-31)_
-```
-
-## Notes
-
-- This command APPLIES fixes, not just logs them
-- Ask questions to understand the real issue
-- Small fixes: edit directly
-- Complex fixes: delegate to Ada
-- Always log what was done for history

package/.opencode/commands/mcp-list.md

@@ -1,27 +0,0 @@
----
-description: List available and enabled MCP servers
----
-
-# List MCP Servers
-
-Show available MCP servers and their status.
-
-## Instructions
-
-1. Read `~/.mcp.json` (global servers) - look for `mcpServers` keys
-2. Read `./.mcp.json` (project servers) if it exists - look for `mcpServers` keys
-3. Read `.claude/settings.local.json` to find enabled servers (`enabledMcpjsonServers` array)
-4. Display a table showing all discovered servers:
-
-| Server | Source | Status |
-|--------|--------|--------|
-| chrome-devtools | global | enabled/disabled |
-| notebooklm | global | enabled/disabled |
-| hailer | project | enabled/disabled |
-
-A server is **enabled** if it's in `enabledMcpjsonServers` array.
-
-Also show:
-- `/mcp-enable <name>` to enable
-- `/mcp-disable <name>` to disable
-- Restart required after changes (`claude -c`)

package/.opencode/commands/onboard-squad.md

@@ -1,140 +0,0 @@
----
-description: Bootstrap a new Hailer project end-to-end
-argument-hint: "project description or PRD path"
-allowed-tools: Task, Bash, Read
----
-# Onboard Squad
-
-Sequential pipeline: Create workspace infrastructure, scaffold the app, and verify everything is wired correctly.
-
-**Agents:**
-1. **Helga** - Workflow config (create workflows, fields, phases, teams, groups)
-2. **Alejandro** - Function fields and nameFunction
-3. **Giuseppe** - Scaffold first app with basic CRUD interface
-4. **Bjorn** - Audit config accuracy and field references
-
-**Goal:** $ARGUMENTS
-
-## Protocol
-
-### Step 1: Parse Requirements
-
-Parse `$ARGUMENTS` to determine input type:
-
-- **If it's a file path** (e.g., `docs/prd-onboarding.md`): Read it for requirements
-- **Otherwise:** Use it as a project description
-
-If `$ARGUMENTS` is a file path, read it first before proceeding to Helga.
-
-### Step 2: Helga (Workspace Config)
-
-Spawn Helga to create workspace infrastructure:
-
-```
-Task(subagent_type="agent-helga-workflow-config", prompt="Bootstrap a new Hailer project: $ARGUMENTS\n\nCreate: workflows (at least 1), fields for each workflow, phases, teams, and groups as needed.")
-```
-
-Wait for result. If Helga returns `"status": "ready_to_push"`:
-- Run the push commands she provides (e.g., `npm run fields-push:force`)
-- Then run `npm run pull` to refresh local workspace with new IDs
-- **Important:** Helga ALWAYS spawns Alejandro for nameFunction. Check if her result indicates Alejandro was spawned and created nameFunction. If yes, Alejandro's work is done - skip to Step 3.
-
-### Step 3: Alejandro (Function Fields) - Conditional
-
-Only spawn if Helga indicates additional function fields are needed beyond what was created by Helga+Alejandro auto-spawn.
-
-If needed, spawn Alejandro:
-
-```
-Task(subagent_type="agent-alejandro-function-fields", prompt="Create additional function fields for workflows created in: [INSERT HELGA'S GOAL].\n\nHelga created these workflows: [SUMMARIZE HELGA'S RESULT - workflow names, IDs, field counts].\n\nFunction fields already created by Helga: [FROM HELGA'S RESULT].")
-```
-
-Wait for result. If `"status": "ready_to_push"`:
-- Run the push commands
-- Run `npm run pull` to refresh workspace
-
-### Step 4: Giuseppe (App Scaffold)
-
-**Before spawning Giuseppe, enable builder mode:**
-```
-Bash: node .claude/hooks/app-edit-guard.cjs --agent-on
-```
-
-Spawn Giuseppe to build the first app using the new workspace:
-
-```
-Task(subagent_type="agent-giuseppe-app-builder", prompt="Scaffold the first Hailer app for this new project: $ARGUMENTS\n\nWorkspace created by Helga: [SUMMARIZE HELGA'S RESULT - workflow names, IDs, field counts]. Giuseppe: read workspace/ directly to discover exact field IDs, types, and team configurations. Do not guess IDs.\n\nBuild a basic CRUD interface using @hailer/app-sdk and Chakra UI. Start with the primary workflow. Use the Hailer Design System.\n\nNote: Assume Local Dev app exists (orchestrator checks this before spawning you). Build for local development at http://localhost:3000.")
-```
-
-**After Giuseppe completes, disable builder mode:**
-```
-Bash: node .claude/hooks/app-edit-guard.cjs --agent-off
-```
-
-Wait for result. If Giuseppe returns `"status": "ready_to_push"`:
-- Run the push commands
-
-### Step 5: Bjorn (Config Audit)
-
-Spawn Bjorn to verify everything is configured correctly:
-
-```
-Task(subagent_type="agent-bjorn-config-audit", prompt="Audit this new Hailer project setup:\n\n1. Verify CLAUDE.md accuracy (check that project type detection works)\n2. Verify workspace configuration: workflows, fields, phases, teams all match workspace/ files\n3. Verify field references in the new app match workspace/ field definitions\n4. Check for missing imports, incorrect IDs, or type mismatches\n5. Verify that nameFunction is set on the primary workflow\n\nProject goal: $ARGUMENTS\n\nReport any misconfigurations. If iteration <= 1, recommend fixes to the orchestrator.")
-```
-
-**Loop Control:**
-- If Bjorn finds no critical issues: proceed to Step 6 (report)
-- If Bjorn finds issues AND `iteration <= 1`:
-  - Increment `iteration`
-  - Determine which agent to fix it (usually Helga or Giuseppe)
-  - Run appropriate agent with the issues to fix
-  - Return to Step 5 (Bjorn re-audits)
-- If Bjorn finds issues AND `iteration > 1`: escalate to user in Step 6
-
-### Step 6: Report
-
-```markdown
-## Onboard Squad Complete
-
-**Project Goal:** $ARGUMENTS
-
-### Workspace Setup (Helga)
-- Workflows created: [list with IDs]
-- Fields created: [count]
-- Phases: [count]
-- Teams: [count]
-- Alejandro auto-spawn: [nameFunction created? Y/N]
-
-### Function Fields (Alejandro)
-- Additional function fields created: [list or "None needed"]
-- Skipped: [reason if skipped]
-
-### App Scaffold (Giuseppe)
-- App path: [path to new app]
-- Build status: Pass/Fail
-- Files created: [count/list]
-- Local dev setup: [check local dev app name]
-
-### Verification (Bjorn)
-- Critical issues found: [count] or "None"
-- Status: VERIFIED / NEEDS FIXES (if iteration > 1, see below)
-
-[If iteration > 1 - escalated:]
-### Remaining Issues
-[List issues Bjorn found that weren't resolved in 1 fix iteration]
-- Manual fixes recommended: [hints]
-
-### Next Steps
-1. Run `npm run dev` to test the app locally
-2. Ensure "Local Dev" app exists in Hailer workspace
-3. Test the app inside Hailer iframe at http://localhost:3000
-4. Run `npm run pull` to ensure workspace/ is fully up to date
-```
-
-## Notes
-
-- Each agent in the pipeline needs IDs from the previous step. Always `npm run pull` between Helga→Alejandro→Giuseppe to refresh local workspace.
-- Helga auto-spawns Alejandro for nameFunction. Check her result - only spawn Alejandro again in Step 3 if additional function fields are needed.
-- Giuseppe needs real workflow and field IDs from workspace/ - always read the files and paste actual IDs, never let him guess.
-- Bjorn's audit is the quality gate. If issues are found and you've iterated once, escalate remaining issues to the user rather than looping infinitely.
-- Set `iteration = 1` at the start. Increment after each Bjorn audit if fixes are attempted.