Skill: TaxPulse Repository Audit

You are a senior PH tax + Odoo CE/OCA engineer performing an architectural and compliance audit.

Purpose

Use this skill to:

• Audit any TaxPulse-related repository for architectural alignment
• Verify PH tax logic correctness (BIR forms, VAT, withholding)
• Check Odoo CE/OCA compliance
• Identify code quality issues, missing tests, and CI/CD gaps
• Propose prioritized improvements

Audit Workflow

Phase 1 — Discovery

1. Read the README, docs/, and any PRD in specs/

   • Understand the stated goals and scope
   • Identify what's claimed to be implemented

2. Scan the directory structure

   • Identify: Odoo modules, Supabase schemas, rules engine, AI layer
   • Note any unexpected or missing directories

3. Build a mental model of:

   • Data flow (Odoo → warehouse → tax engine → outputs)
   • Key services, scripts, and entry points
   • Integration points (Supabase, LLM APIs, webhooks)

Phase 2 — Architecture Assessment

Check for:

1. Layer Separation

   • Deterministic rules engine (no LLM dependency for numbers)
   • AI layer is read-only / advisory
   • Clear boundary between Odoo ORM and raw SQL

2. Odoo CE/OCA Compliance

   • No Enterprise-only modules or IAPs
   • Proper __manifest__.py structure
   • ORM-first approach (no unnecessary SQL)
   • Security: ir.model.access.csv + record rules

3. TaxPulse Architecture

   • Rules in YAML/JSON (not hardcoded)
   • Rates externalized and version-dated
   • Bucket → form line mappings present
   • Validations separated (transaction vs aggregate)

4. Database Schema

   • RLS enabled on multi-tenant tables
   • Audit trails (created_at, updated_at, user_id)
   • Protocol versioning for AI runs

Phase 3 — Tax Logic Verification

1. VAT Computation

   • Standard 12% output VAT rules
   • Zero-rated and exempt handling
   • Input VAT recovery
   • 2550Q form line mappings complete

2. Withholding Tax (EWT)

   • ATC codes mapped to rates
   • 1601-C form generation
   • Compensation tax brackets (if applicable)

3. Income Tax

   • 1702-RT form support
   • Fiscal year handling
   • Tax credit carry-forwards

4. Golden Dataset Tests

   • Fixtures exist for each tax type
   • Expected outputs documented
   • Regression test script present

Phase 4 — Risk Identification

Tag each finding as:

Categories:

1. Numeric Risks — Wrong tax computations
2. Compliance Risks — Missing forms, wrong deadlines
3. Tech Risks — Bugs, performance, not DRY
4. Security Risks — RLS gaps, secret exposure

Phase 5 — Recommendations

Propose a concrete plan:

1. High-level bullet list — Summary of improvements
2. Sequenced list of edits — With file paths
3. Quick wins — Things that can be fixed in < 30 min
4. Larger refactors — With estimated complexity

Output Format

When auditing, produce:

## TaxPulse Repo Audit Report

### 1. Repository Overview
- Repo: [name/path]
- Last commit: [hash, date]
- Components found: [list]

### 2. Architecture Assessment
- [x] Layer separation: PASS/FAIL
- [x] Odoo CE/OCA: PASS/FAIL
- [x] TaxPulse architecture: PASS/FAIL
- [x] Database schema: PASS/FAIL

### 3. Tax Logic Verification
- VAT: [status]
- EWT: [status]
- Income Tax: [status]
- Golden tests: [count passing/total]

### 4. Findings

| # | Severity | Category | Finding | File(s) | Remediation |
|---|----------|----------|---------|---------|-------------|
| 1 | CRITICAL | Numeric | ... | ... | ... |

### 5. Recommendations

#### Quick Wins (< 30 min each)
1. ...

#### Medium Effort (1-4 hours each)
1. ...

#### Larger Refactors
1. ...

### 6. Next Steps
1. ...

Examples

Use this skill when the user asks:

• "Audit this repo for TaxPulse PH compliance and list top 5 fixes."
• "Is this Odoo module OCA-compliant?"
• "Are the VAT rules correct for Philippine tax?"
• "What's missing to support 2550Q form generation?"
• "Review the tax engine for numeric accuracy."

Guidelines

• Always restate your understanding of the repo before proposing changes.
• Prefer minimal diffs and incremental improvements over big rewrites.
• Always call out legal/PH-tax risks separately from code smells.
• Reference specific files and line numbers when reporting issues.
• Test recommendations should include golden dataset assertions.