feat(gtm): add mandatory knowledge-base read/write directives to GTM skills

Agents must now read client profiles before GTM work and write session
logs after completing audits, tag changes, or validation — ensuring
findings compound across sessions instead of being lost.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

.claude/commands/gtm-audit.md

@@ -6,6 +6,20 @@ description: GTM page audit - scan fired tags, gap analysis, tag design from DOM
 
 Comprehensive Google Tag Manager audit using Playwright to scan live pages for container health, tag firing, dataLayer events, form tracking, and e-commerce checkout flows.
 
+## MANDATORY: Knowledge Base Read/Write
+
+**Before starting any audit:**
+1. Identify the target client from container ID or URL
+2. Read `knowledge-base/accounts/<client>/profile.md` — URL patterns, known issues, platform stack, past findings
+3. Read `knowledge-base/accounts/<client>/*.md` — taxonomy, naming issues to exclude from new findings
+4. Skim `knowledge-base/logs/<client>/` — past session context
+
+**After completing audit, write a session log:**
+- Write to `knowledge-base/logs/<client>/YYYY-MM-DD-<description>.md`
+- Include: date, container ID, status, issues found (root causes + specific IDs), lessons learned
+- Update `knowledge-base/accounts/<client>/profile.md` if you discovered persistent facts
+- See `AGENTS.md` for full format template
+
 ## Triggers
 - "audit GTM", "check dataLayer", "GTM 검사", "scan GTM tags", "audit tags on page", "check tag firing"
 

.claude/commands/gtm-editor.md

@@ -22,6 +22,20 @@ Priority reads per task:
 - "create GTM tag", "generate dataLayer", "modify trigger"
 - "update variable", "write custom HTML", "manage GTM"
 
+## MANDATORY: Knowledge Base Read/Write
+
+**Before creating or modifying ANY tags, triggers, or variables:**
+1. Identify the target client from container ID
+2. Read `knowledge-base/accounts/<client>/profile.md` — key variables, URL patterns, platform stack, known issues
+3. Read `knowledge-base/accounts/<client>/*.md` — taxonomy reveals existing event names and naming conventions
+4. Skim `knowledge-base/logs/<client>/` — past fixes reveal patterns (e.g., "form POST loses URL params")
+
+**After completing tag management, write a session log:**
+- Write to `knowledge-base/logs/<client>/YYYY-MM-DD-<description>.md`
+- Include: date, container ID, tags affected (IDs), what was created/modified/deleted, cHTML snippets
+- Update `knowledge-base/accounts/<client>/profile.md` if you discovered persistent facts
+- See `AGENTS.md` for full format template
+
 ## MANDATORY: dataLayer First Workflow
 
 1. **Design dataLayer push FIRST** (match client's tech stack: vanilla JS/React/Vue/PHP)

.claude/commands/gtm-validator.md

@@ -20,6 +20,20 @@ Priority reads per task:
 - "validate tags", "QA GTM", "debug GTM"
 - "naming conventions", "GTM best practice"
 
+## MANDATORY: Knowledge Base Read/Write
+
+**Before starting any validation or QA:**
+1. Identify the target client from container ID or URL
+2. Read `knowledge-base/accounts/<client>/profile.md` — naming conventions, known issues, platform stack
+3. Read `knowledge-base/accounts/<client>/*.md` — taxonomy defines expected events; naming fix plans reveal known violations
+4. Skim `knowledge-base/logs/<client>/` — past QA reveals recurring issues and known false positives
+
+**After completing validation, write a session log:**
+- Write to `knowledge-base/logs/<client>/YYYY-MM-DD-<description>.md`
+- Include: date, container ID, pass/fail results, broken triggers (IDs), new naming violations
+- Update `knowledge-base/accounts/<client>/profile.md` if you discovered persistent facts
+- See `AGENTS.md` for full format template
+
 ## Validation Modes
 
 ### 1. Tag Firing Verification