Agent Reconciliation

Agent Reconciliation Skill

Overview

This skill enables systematic reconciliation between the filesystem agent definitions (AGENTS.md files under docs-paperclip/companies/*/agents/) and the database tables that power the Paperclip control plane (agents, agent_models, agent_api_keys).

Trigger Conditions

• User asks to "reconcile agents"
• User asks to "check agents against database"
• User asks to "verify agent registrations"
• User asks to "audit agent names" or "check funky human names"
• User asks to "validate naming conventions"
• User asks to "find missing agents" or "find extra agents"
• User asks to "generate agent fix SQL"

Execution Workflow

Fresh Reconciliation Workflow (Recommended for Full Sync)

When doing a complete sync of all agents from filesystem to database:

1. Run fresh_reconciliation.py

   python3 fresh_reconciliation.py
   

   This will:

   • Scan all filesystem agents
   • Generate fresh_reconciliation.sql (DELETE + INSERT statements)
   • Generate fresh_reconciliation_summary.json
   • Warn that it will DELETE existing data

2. Review Generated SQL

   • Open fresh_reconciliation.sql
   • Verify DELETE statement targets correct tables
   • Verify INSERT statements have correct column values
   • Check for any unexpected DELETE operations

3. Execute Against Database (with user approval)

   # Apply the SQL (this DELETES all existing agents first!)
   psql $DATABASE_URL -f fresh_reconciliation.sql
   

4. Post-Execution Verification

   • Query agents table count
   • Verify sample agents have correct names/titles
   • Check agent_api_keys were created
   • Verify agent_models assignments

Step 0: Schema Validation (MANDATORY)

Before generating ANY SQL, you MUST validate the schema:

1. Run CHECK-column-existence.sql template from docs-paperclip/schema/templates/

   SELECT column_name, data_type 
   FROM information_schema.columns 
   WHERE table_name = 'agents' 
   ORDER BY ordinal_position;
   

2. Verify template compatibility - Compare results against expected columns in template header comments

3. CRITICAL: The agents table has TWO name fields:

   • name = human-readable name (e.g., "SEO Strategist", "DB Specialist")
   • title = agent identifier/slug (e.g., "contentforge-ai-SEOStrategist")

   DO NOT confuse name with slug - slug does NOT exist in the schema!

   CORRECT MAPPING:

   -- ✅ CORRECT:
   INSERT INTO agents (name, title, ...) VALUES ('SEO Strategist', 'contentforge-ai-SEOStrategist', ...);
   
   -- ❌ WRONG - name should be human-readable, not the slug:
   INSERT INTO agents (name, title, ...) VALUES ('contentforge-ai-SEOStrategist', 'SEO Strategist', ...);
   

Step 1: Filesystem Discovery

1. Scan docs-paperclip/companies/{company}/agents/**/AGENTS.md
2. Parse YAML frontmatter for each agent
3. Extract: name, slug, reportsTo, team, skills
4. Record directory name for naming convention checks
5. Build filesystem agent index

Step 2: Database Discovery

1. Query companies table → map name to id
2. Query agents table → all rows with id, name, title, company_id, role, reports_to, status
3. Query agent_models table → active model assignments by agent_id (TEXT type)
4. Query agent_api_keys table → API keys by agent_id (UUID type)
5. Build database agent index

Step 3: Reconciliation Match

1. Match filesystem agents to database agents by name ↔ name
2. Identify missing agents: in filesystem, not in database
3. Identify extra agents: in database, not in filesystem
4. Identify agents without API keys
5. Identify agents without model assignments
6. Identify invalid human names: title ≥23 characters or invalid characters
7. Identify naming convention violations: directory doesn't follow {company-slug}-{AgentName}

Step 4: SQL Fix Generation (Using Templates)

ALWAYS use templates from docs-paperclip/schema/templates/ as the source of truth.

IMPORTANT: For agent title/human name fixes, use title NOT slug:

-- ✅ CORRECT:
UPDATE agents SET title = 'Procurement' WHERE name = 'domainforge-ai-domainforge-ai-procurement';

-- ❌ WRONG - 'slug' column does NOT exist:
UPDATE agents SET slug = 'Procurement' WHERE ...

Step 5: Collaboration Delegation

• Complex schema queries → delegate to paperclipforge-ai-database-druid
• Agent creation logic → delegate to paperclipforge-ai-atlas-agent-creator
• Model assignment logic → delegate to paperclipforge-ai-model-assignment-specialist

Step 6: Report Output

• Save SQL to reconciliation_fixes.sql
• Save JSON report to reconciliation_report.json
• Print summary with counts per company

SQL Template Location

All SQL generation MUST use templates from:

docs-paperclip/schema/templates/

Template files:

• README.md - Template registry documentation
• CHECK-column-existence.sql - Schema validation (run FIRST)
• VALIDATE-agent-exists.sql - Pre-modification validation
• REGISTER-agent.sql - Create new agents
• UPDATE-agent-title.sql - Fix human names (title field)
• DELETE-agent.sql - Safe deletion with FK cascade
• ADD-agent-api-keys.sql - Add API keys

Single Source of Truth

The authoritative schema is in packages/db/src/schema/*.ts. Templates are validated against this source. If template columns don't match actual schema, run CHECK-column-existence.sql and update template accordingly.

Naming Convention Rules

Human Name (Canonical Location: AGENTS.md YAML frontmatter)

• Field: human_name in AGENTS.md YAML frontmatter
• Must be <23 characters
• Allowed characters: a-z, A-Z, 0-9, spaces, hyphens, underscores
• Should be human-readable (e.g., "Nexus", "Dataforge", "Recon")
• Reference: See docs-paperclip/companies/shared/AGENT-NAMING-STANDARD.md

Directory Name

• Pattern: {company-slug}-{HumanNameWithoutSpaces}
• Examples: devforge-ai-Nexus, measureforge-ai-CadOrchestrator
• Must match the company slug prefix
• Folder suffix should match human_name with spaces removed

File Name

• Always AGENTS.md
• Always inside the agent directory

Directory Name Alignment Workflow

When user asks to "align folder names with human names" or "rename agent folders":

Step 1: Scan All Agents

1. Scan docs-paperclip/companies/*/agents/**/AGENTS.md
2. Extract human_name from YAML frontmatter
3. Record current folder name

Step 2: Identify Mismatches

For each agent:

• Expected folder: {company-slug}-{human_name_without_spaces}
• If current folder doesn't match expected → flag for rename

Step 3: Generate Rename Commands

mv "old-folder-name" "expected-folder-name"

Step 4: Execute Renames (with user confirmation)

• Present dry-run first showing all planned renames
• Execute only after user approval

Step 5: Update register-agent.sql files

After folder rename, update any register-agent.sql files to reference new folder path.

Human Name Standardization Workflow

When user asks to "standardize human names" or "fix agent human names":

Step 1: Scan All Agents

1. Find all AGENTS.md files
2. Extract current human_name values
3. Identify issues: missing, truncated, >22 chars

Step 2: Fix Issues

• Missing: Derive from folder name (convert CamelCase to spaces)
• Truncated: Expand to full meaningful name (max 22 chars)
• Too long: Truncate intelligently, preserve key words

Step 3: Validate

• All names <23 characters
• All names human-readable
• All names match folder suffix pattern

Input Parameters

Output Format

Console Summary

COMPANIES:
  - In filesystem: 16
  - In database: 16
AGENTS:
  - In filesystem: 496
  - In database: 548
  - Missing in database: 344
  - Extra in database: 402
  - Without API keys: 84
  - Without models: 159
VALIDATION ISSUES:
  - Invalid names: 70
  - Naming convention issues: 0

JSON Report Structure

{
  "companies_in_filesystem": [...],
  "companies_in_database": [...],
  "missing_companies": [...],
  "missing_agents": [{"name": "...", "slug": "...", "company": "..."}],
  "extra_agents_in_database": [{"id": "...", "name": "..."}],
  "agents_without_api_keys": [{"id": "...", "name": "..."}],
  "agents_without_models": [{"id": "...", "name": "..."}],
  "invalid_names": [{"name": "...", "reason": "..."}],
  "naming_convention_issues": [...],
  "sql_count": 657
}

Error Handling

• Missing companies directory: Report error and exit
• Database connection failure: Report error with connection details
• Malformed AGENTS.md: Log warning, skip file, continue
• No database agents: Report warning but continue with filesystem scan

Error Logging for LearningForge Integration

IMPORTANT: All errors MUST be logged to activity_log for LearningForge AI error pattern monitoring.

Required Error Logging

When any of these errors occur, log to activity_log with proper error categorization:

-- Log SQL validation errors
INSERT INTO activity_log (
  id, company_id, actor_type, actor_id, action, entity_type, entity_id, details, created_at
) VALUES (
  gen_random_uuid(),
  (SELECT id FROM companies WHERE name = 'PaperclipForge AI'),
  'agent',
  'paperclipforge-ai-agent-reconciler',
  'error',
  'agent',
  NULL,
  '{
    "error_type": "sql_validation",
    "error_message": "Column slug does not exist in agents table",
    "context": "reconciliation SQL generation",
    "remediation": "Use title field instead of slug"
  }'::jsonb,
  NOW()
);

Error Categories for Logging

Integration with LearningForge AI

The learningforge-ai-error-pattern-monitor agent monitors activity_log for these error patterns and creates alerts when thresholds are exceeded.

Skill Name: agent-reconciliation Version: 1.0 Owner: paperclipforge-ai-agent-reconciler Created: 2026-04-24