oh-my-customcodex 0.1.8 → 0.2.0

package/README.md

@@ -13,7 +13,7 @@
 
 **[한국어 문서 (Korean)](./README_ko.md)**
 
-48 agents. 108 skills. 22 rules. One command.
+48 agents. 112 skills. 22 rules. One command.
 
 ```bash
 npm install -g oh-my-customcodex && cd your-project && omcodex init
@@ -132,14 +132,14 @@ Each agent declares its tools, model, memory scope, and limitations in YAML fron
 
 ---
 
-### Skills (108)
+### Skills (112)
 
 | Category | Count | Includes |
 |----------|-------|----------|
 | Best Practices | 24 | Go, Python, TypeScript, Kotlin, Rust, React, FastAPI, Spring Boot, Django, Flutter, Docker, AWS, Postgres, Redis, Kafka, dbt, Spark, Snowflake, Airflow, pipeline-architecture-patterns, alembic, and more |
 | Routing | 4 | secretary, dev-lead, de-lead, qa-lead |
 | Workflow | 13 | structured-dev-cycle, deep-plan, research, evaluator-optimizer, dag-orchestration, worker-reviewer-pipeline, reasoning-sandwich, pipeline, and more |
-| Development | 9 | dev-review, dev-refactor, analysis, create-agent, intent-detection, web-design-guidelines, omcodex:takeover, skill-extractor, idea |
+| Development | 10 | dev-review, dev-refactor, analysis, create-agent, intent-detection, web-design-guidelines, omcodex:takeover, skill-extractor, pre-generation-arch-check, idea |
 | Operations | 10 | update-docs, audit-agents, sauron-watch, monitoring-setup, token-efficiency-audit, fix-refs, release-notes, and more |
 | Memory | 3 | memory-save, memory-recall, memory-management |
 | Package | 3 | npm-publish, npm-version, npm-audit |
@@ -166,6 +166,7 @@ All commands are invoked inside the oh-my-customcodex GPT Codex + OMX session.
 | `/research` | 10-team parallel analysis with cross-verification |
 | `/sdd-dev` | Spec-Driven Development workflow |
 | `/ambiguity-gate` | Pre-routing ambiguity analysis |
+| `/pre-generation-arch-check` | Check architecture risks before implementation |
 | `/adversarial-review` | Attacker-mindset security code review |
 | `/pipeline` | Execute YAML-defined pipelines |
 | `/pipeline resume` | Resume a halted pipeline from last failure point |
@@ -224,7 +225,7 @@ Key rules: R010 (orchestrator never writes files), R009 (parallel execution mand
 
 ---
 
-### Guides (38)
+### Guides (39)
 
 Reference documentation covering best practices, architecture decisions, and integration patterns. Located in `guides/` at project root, covering topics from agent design to CI/CD to observability.
 
@@ -281,7 +282,7 @@ your-project/
 │   ├── specs/                  # Extracted canonical specs
 │   ├── contexts/               # 4 shared context files
 │   └── ontology/               # Knowledge graph for RAG
-└── guides/                     # 38 reference documents
+└── guides/                     # 39 reference documents
 ```
 
 ---

package/dist/cli/index.js

@@ -3087,7 +3087,7 @@ var init_package = __esm(() => {
     workspaces: [
       "packages/*"
     ],
-    version: "0.1.8",
+    version: "0.2.0",
     description: "Batteries-included agent harness on top of GPT Codex + OMX",
     type: "module",
     bin: {

package/dist/index.js

@@ -2173,7 +2173,7 @@ var package_default = {
   workspaces: [
     "packages/*"
   ],
-  version: "0.1.8",
+  version: "0.2.0",
   description: "Batteries-included agent harness on top of GPT Codex + OMX",
   type: "module",
   bin: {

package/package.json

@@ -3,7 +3,7 @@
   "workspaces": [
     "packages/*"
   ],
-  "version": "0.1.8",
+  "version": "0.2.0",
   "description": "Batteries-included agent harness on top of GPT Codex + OMX",
   "type": "module",
   "bin": {

package/templates/.claude/hooks/hooks.json

@@ -336,6 +336,16 @@
       }
     ],
     "PostToolUse": [
+      {
+        "matcher": "tool matches \"^mcp__playwright__\"",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "bash .codex/hooks/scripts/playwright-compress.sh"
+          }
+        ],
+        "description": "Compress verbose Playwright MCP output while preserving refs and URLs"
+      },
       {
         "matcher": "tool == \"Bash\"",
         "hooks": [

package/templates/.claude/hooks/scripts/playwright-compress.sh

@@ -0,0 +1,79 @@
+#!/bin/bash
+set -euo pipefail
+
+input=$(cat)
+tool_name=$(echo "$input" | jq -r '.tool_name // .tool // ""')
+
+if [[ ! "$tool_name" =~ ^mcp__playwright__ ]]; then
+  exit 0
+fi
+
+raw_output=$(echo "$input" | jq -c '.tool_response // .tool_output.output // .tool_output // empty')
+
+if [ -z "$raw_output" ] || [ "$raw_output" = "null" ]; then
+  exit 0
+fi
+
+text_output=$(printf '%s' "$raw_output" | jq -r 'if type == "string" then . else tostring end')
+min_chars=${PLAYWRIGHT_COMPRESS_MIN_CHARS:-3000}
+
+if [ "${#text_output}" -lt "$min_chars" ]; then
+  exit 0
+fi
+
+ref_matches=$(printf '%s\n' "$text_output" | grep -oE 'ref[=:]"?[^", ]+' || true)
+refs=$(printf '%s\n' "$ref_matches" \
+  | sed 's/^ref[=:]"*/ref=/' \
+  | awk 'length($0) > 0 && !seen[$0]++' \
+  | head -12 \
+  | paste -sd ', ' -)
+
+url_matches=$(printf '%s\n' "$text_output" | grep -oE 'https?://[^" )]+' || true)
+urls=$(printf '%s\n' "$url_matches" \
+  | awk 'length($0) > 0 && !seen[$0]++' \
+  | head -6 \
+  | paste -sd ', ' -)
+
+important_matches=$(printf '%s\n' "$text_output" | grep -iE 'ref=|error|failed|warning|timeout|assert|url|title|text|name|role|selector' || true)
+important_lines=$(printf '%s\n' "$important_matches" \
+  | sed 's/^[[:space:]]*//; s/[[:space:]]\+/ /g' \
+  | awk 'length($0) > 0 && !seen[$0]++' \
+  | head -25)
+
+if [ -z "$important_lines" ]; then
+  important_lines=$(printf '%s\n' "$text_output" \
+    | sed 's/^[[:space:]]*//; s/[[:space:]]\+/ /g' \
+    | awk 'length($0) > 0 && !seen[$0]++' \
+    | head -20)
+fi
+
+if [ -z "$important_lines" ]; then
+  important_lines='(no high-signal lines detected; payload was compressed by fallback rules)'
+fi
+
+summary=$(
+  {
+    printf 'Playwright MCP output compressed.\n'
+    printf 'Original size: %s chars.\n' "${#text_output}"
+    if [ -n "$refs" ]; then
+      printf 'Preserved refs: %s\n' "$refs"
+    fi
+    if [ -n "$urls" ]; then
+      printf 'URLs: %s\n' "$urls"
+    fi
+    printf '\nHigh-signal lines:\n'
+    printf '%s\n' "$important_lines"
+  } | head -c "${PLAYWRIGHT_COMPRESS_MAX_CHARS:-2500}"
+)
+
+if [ -z "$summary" ]; then
+  exit 0
+fi
+
+jq -n \
+  --arg summary "$summary" \
+  --arg note "Playwright MCP output compressed to preserve refs and reduce context cost." \
+  '{
+    additionalContext: $note,
+    updatedMCPToolOutput: $summary
+  }'

package/templates/.claude/rules/SHOULD-ecomode.md

@@ -18,6 +18,8 @@ Auto-activates when: 4+ parallel tasks, batch operations, 80%+ context usage, or
 
 **Compression**: File lists -> count only (unless < 5), error traces -> first/last 3 lines, code -> path:line ref only.
 
+**Pruning Transparency**: When input pruning occurs, include a compact report line such as `[Pruned] {n} files, ~{tokens} tokens saved` so the user can see what was compressed away.
+
 ## Config
 
 ```yaml

package/templates/.claude/skills/design-shotgun/SKILL.md

@@ -0,0 +1,72 @@
+---
+name: design-shotgun
+description: Generate and compare multiple deliberate UI directions before converging on one implementation path
+scope: core
+user-invocable: true
+argument-hint: "<screen-or-flow>"
+---
+
+# Design Shotgun
+
+Use this skill when one polished mockup is too early and the real need is breadth first. The goal is not random variety; it is structured divergence followed by deliberate convergence.
+
+## When To Use
+
+- early UI exploration
+- redesign of a high-visibility screen
+- style selection before implementation
+- multiple viable directions need comparison
+
+## Workflow
+
+### 1. Pick comparison axes
+
+Choose 2-3 axes that should differ between options:
+
+- editorial vs. utilitarian
+- dense vs. spacious
+- bold vs. restrained
+- product-led vs. workflow-led
+
+### 2. Generate 4-6 directions
+
+Each direction should differ materially in:
+
+- hierarchy
+- composition
+- copy tone
+- motion or interaction emphasis
+
+Avoid trivial color-swaps.
+
+### 3. Compare on one board
+
+For each direction, record:
+
+- strongest move
+- weakest move
+- what kind of user it favors
+- what should survive into the final hybrid
+
+### 4. Build taste memory
+
+Preserve:
+
+- chosen typography moves
+- preferred spacing rhythm
+- colors or motifs worth keeping
+- rejected patterns to avoid repeating
+
+### 5. Converge
+
+Produce one implementation direction with:
+
+- headline design thesis
+- components to build first
+- references to related guides or skills
+
+## Related References
+
+- `.codex/skills/impeccable-design/SKILL.md`
+- `.codex/skills/web-design-guidelines/SKILL.md`
+- `guides/browser-automation/README.md` for consistent browser-based comparison capture

package/templates/.claude/skills/intent-detection/patterns/agent-triggers.yaml

@@ -307,6 +307,30 @@ agents:
   # ---------------------------------------------------------------------------
   # Research / Information Gathering (skill-based, not agent)
   # ---------------------------------------------------------------------------
+  wiki-rag:
+    keywords:
+      korean:
+        - 아키텍처
+        - 워크플로우
+        - 어떤 에이전트
+        - 어떤 스킬
+        - 규칙
+      english:
+        - architecture
+        - workflow
+        - "which agent"
+        - "which skill"
+        - rule
+        - "how does this work"
+    file_patterns: []
+    supported_actions:
+      - explain
+      - search
+      - investigate
+      - research
+    base_confidence: 80
+    routing_rule: "Use wiki-rag for architecture, workflow, agent, skill, and rule questions before broader exploration"
+
   research-workflow:
     keywords:
       korean:
@@ -369,6 +393,90 @@ agents:
     action_weight: 25
     routing_rule: "Suggest codex-exec hybrid when codex available, or gemini-exec hybrid when gemini available, for new file creation tasks"
 
+  product-strategy:
+    keywords:
+      korean:
+        - 제품 전략
+        - 오피스 아워
+        - 창업자 질문
+        - CEO 리뷰
+        - 범위 축소
+      english:
+        - "product strategy"
+        - "office hours"
+        - "ceo review"
+        - "founder questions"
+        - "scope reduction"
+    file_patterns: []
+    supported_actions:
+      - analyze
+      - plan
+      - review
+      - reduce
+    base_confidence: 65
+    routing_rule: "Use the product-strategy skill to pressure-test product bets before implementation"
+
+  design-shotgun:
+    keywords:
+      korean:
+        - 디자인 샷건
+        - 시안 비교
+        - 테이스트 메모리
+        - 방향 비교
+      english:
+        - "design shotgun"
+        - "compare directions"
+        - "taste memory"
+        - "variation board"
+    file_patterns: []
+    supported_actions:
+      - create
+      - compare
+      - review
+      - polish
+    base_confidence: 65
+    routing_rule: "Use the design-shotgun skill for multi-direction UI exploration before converging on one implementation"
+
+  browser-automation-reference:
+    keywords:
+      korean:
+        - 브라우저 자동화
+        - 쿠키 임포트
+        - 안티봇
+        - 세션 재사용
+      english:
+        - "browser automation guide"
+        - "cookie import"
+        - anti-bot
+        - "session reuse"
+    file_patterns: ["playwright.config.ts", "*.pw.ts"]
+    supported_actions:
+      - explain
+      - review
+      - investigate
+    base_confidence: 55
+    routing_rule: "Consult guides/browser-automation/README.md and existing Playwright surfaces before implementing browser-session workflows"
+
+  playwright-compress:
+    keywords:
+      korean:
+        - 플레이라이트 압축
+        - 브라우저 출력 압축
+        - ref 보존
+        - MCP 압축
+      english:
+        - "playwright compress"
+        - "browser output compression"
+        - "preserve refs"
+        - "mcp compression"
+    file_patterns: ["playwright.config.ts", "*.pw.ts"]
+    supported_actions:
+      - optimize
+      - compress
+      - review
+    base_confidence: 60
+    routing_rule: "Use the playwright-compress skill when Playwright MCP output is verbose and refs must remain actionable"
+
   # Managers (continued)
   mgr-gitnerd:
     keywords:

package/templates/.claude/skills/playwright-compress/SKILL.md

@@ -0,0 +1,46 @@
+---
+name: playwright-compress
+description: Compress verbose Playwright MCP output while preserving element refs and actionable browser evidence
+scope: core
+user-invocable: true
+argument-hint: "[status]"
+---
+
+# Playwright Compress
+
+This skill documents the Codex-native pattern for shrinking verbose Playwright MCP output without destroying the references needed for follow-up interaction.
+
+## Purpose
+
+- keep `ref=` tokens visible
+- reduce repetitive browser snapshots and logs
+- preserve URLs, errors, and key visible text
+- complement runtime token controls instead of replacing them
+
+## Runtime Surface
+
+The implementation lives in:
+
+- `.codex/hooks/scripts/playwright-compress.sh`
+- `.codex/hooks/hooks.json`
+
+It runs on `PostToolUse` for Playwright MCP tools and returns `updatedMCPToolOutput` when the payload is large enough to benefit from compression.
+
+## Compression Policy
+
+- Leave short outputs untouched.
+- Preserve `ref=` tokens and URLs.
+- Prefer high-signal lines over bulk DOM noise.
+- Keep failure evidence verbatim where possible.
+- Never send browser session secrets to external services.
+
+## Use Cases
+
+- browser snapshot output is too large to keep in context
+- a page contains many repeated nodes but only a few actionable refs
+- follow-up tool calls need a compact view of visible evidence
+
+## Related References
+
+- `guides/claude-code/14-token-efficiency.md`
+- `guides/browser-automation/README.md`

package/templates/.claude/skills/pre-generation-arch-check/SKILL.md

@@ -0,0 +1,80 @@
+---
+name: pre-generation-arch-check
+description: Check planned code changes for architecture and responsibility violations before implementation
+scope: package
+argument-hint: "<change-request-summary>"
+user-invocable: true
+---
+
+# Pre-Generation Architecture Check
+
+Review a requested change before code generation begins and warn about likely architecture violations.
+
+This skill fills a gap between planning and implementation:
+
+- `adversarial-review` focuses on attacker-minded risk after code exists
+- `deep-verify` focuses on release-quality verification after code exists
+- `pre-generation-arch-check` focuses on architecture hygiene before code is written
+
+## What It Checks
+
+The skill looks for likely violations of:
+
+- R006 separation of concerns
+- compilation metaphor integrity
+- wrong-layer ownership
+- accidental orchestrator responsibility expansion
+- speculative wrappers or abstractions that do not earn their cost
+
+## Inputs
+
+Provide a short request summary, optionally including likely files or modules.
+
+Examples:
+
+```text
+/pre-generation-arch-check add a background sync service to updater.ts
+/pre-generation-arch-check move routing logic into the agent file instead of the routing skill
+```
+
+## Output Contract
+
+### No concern
+
+```text
+[ARCH-CHECK] CLEAR
+
+Request appears consistent with current architecture.
+```
+
+### Warning
+
+```text
+[ARCH-WARNING] Severity: MEDIUM
+Boundary: routing vs agent definition
+Concern: Request would move reusable routing logic into an agent artifact
+Why: Violates compilation metaphor and R006 separation of concerns
+Safer shape: Keep logic in a skill or routing layer; keep agent file declarative
+```
+
+## Heuristics
+
+- warn when a request mixes skill logic with agent artifacts
+- warn when orchestrator responsibilities expand into direct file-writing logic
+- warn when a request introduces a new abstraction without a clear reuse or boundary win
+- warn when a change smells like the wrong layer owns the behavior
+- prefer concise warnings with one safer alternative
+
+## Integration
+
+Use before:
+
+- major refactors
+- new workflow or routing features
+- multi-file cross-layer changes
+
+Good pairings:
+
+- `pre-generation-arch-check` -> `deep-plan`
+- `pre-generation-arch-check` -> `structured-dev-cycle`
+- `pre-generation-arch-check` -> implementation

package/templates/.claude/skills/product-strategy/SKILL.md

@@ -0,0 +1,63 @@
+---
+name: product-strategy
+description: YC-style product pressure test and scope framing for ideas, features, and workflow bets
+scope: core
+user-invocable: true
+argument-hint: "<idea-or-feature>"
+---
+
+# Product Strategy
+
+Use this skill when a request needs product pressure, not immediate implementation. It borrows the useful part of the "office hours" pattern: force the conversation to answer hard product questions before work expands.
+
+## When To Use
+
+- feature framing
+- product tradeoff review
+- CEO / founder style pressure test
+- scope negotiation before implementation
+
+## Workflow
+
+### 1. State the bet
+
+Write the feature or idea in one sentence:
+
+- who it is for
+- what changes for them
+- why it matters now
+
+### 2. Run the six questions
+
+Answer these before moving forward:
+
+1. What user pain becomes meaningfully smaller?
+2. Why will the user notice this quickly?
+3. What simpler alternative did we reject?
+4. What metric or behavior would prove this worked?
+5. What gets harder if we ship this?
+6. What would we deliberately not build in v1?
+
+### 3. Choose a scope mode
+
+Assign one mode:
+
+- `Expand` — increase ambition because the upside justifies it
+- `Selective` — keep only the highest-signal slice
+- `Hold` — wait for a prerequisite or stronger evidence
+- `Reduce` — cut scope because the current ask is inflated
+
+### 4. Output contract
+
+Return:
+
+- problem statement
+- strongest argument for shipping
+- strongest argument against shipping
+- chosen scope mode
+- next 2-3 concrete actions
+
+## Related References
+
+- `guides/browser-automation/README.md` for grounded product review through real browser flows
+- `.codex/skills/deep-plan/SKILL.md` for implementation planning once the strategy is accepted

package/templates/CLAUDE.md

@@ -104,7 +104,7 @@ oh-my-customcodex로 구동됩니다.
 | 릴리즈 | `/pipeline auto-dev`, `/omcodex-release-notes`, `/release-plan` | 자동 개발, 릴리즈 노트 |
 | 리서치 | `/research`, `/scout`, `/deep-plan`, `/omcodex:agora` | 병렬 분석, URL 평가, 연구 계획 |
 | 메모리 | `/memory-save`, `/memory-recall` | 세션 메모리 관리 |
-| 시스템 | `/token-efficiency-audit`, `/omcodex:lists`, `/omcodex:status`, `/omcodex:help` | 토큰 효율 감사, 전체 목록, 상태, 도움말 |
+| 시스템 | `/token-efficiency-audit`, `/pre-generation-arch-check`, `/omcodex:lists`, `/omcodex:status`, `/omcodex:help` | 토큰 효율/아키텍처 감사, 전체 목록, 상태, 도움말 |
 
 > 전체 커맨드 목록 (60+ 커맨드): `/omcodex:lists` 실행
 
@@ -115,7 +115,7 @@ project/
 +-- AGENTS.md                    # 진입점
 +-- .codex/
 |   +-- agents/                  # 서브에이전트 정의 (48 파일)
-|   +-- skills/                  # 스킬 (108 디렉토리)
+|   +-- skills/                  # 스킬 (109 디렉토리)
 |   +-- rules/                   # 전역 규칙 (R000-R022)
 |   +-- hooks/                   # 훅 스크립트 (보안, 검증, HUD)
 |   +-- contexts/                # 컨텍스트 파일 (ecomode)

package/templates/CLAUDE.md.en

@@ -118,6 +118,7 @@ NO EXCEPTIONS. NO EXCUSES.
 | `/optimize-report` | Generate optimization report |
 | `/research` | 10-team parallel deep analysis and cross-verification |
 | `/deep-plan` | Research-validated planning (research → plan → verify) |
+| `/pre-generation-arch-check` | Check architecture risks before implementation |
 | `/omcodex:sauron-watch` | Full R017 verification |
 | `/structured-dev-cycle` | 6-stage structured development cycle (Plan → Verify → Implement → Verify → Compound → Done) |
 | `/omcodex:lists` | Show all available commands |
@@ -131,7 +132,7 @@ project/
 +-- AGENTS.md                    # Entry point
 +-- .codex/
 |   +-- agents/                  # Subagent definitions (48 files)
-|   +-- skills/                  # Skills (108 directories)
+|   +-- skills/                  # Skills (109 directories)
 |   +-- rules/                   # Global rules (22 files)
 |   +-- hooks/                   # Hook scripts (security, validation, HUD)
 |   +-- contexts/                # Context files (4 files)

package/templates/CLAUDE.md.ko

@@ -118,6 +118,7 @@ oh-my-customcodex로 구동됩니다.
 | `/optimize-report` | 최적화 리포트 생성 |
 | `/research` | 10-team 병렬 딥 분석 및 교차 검증 |
 | `/deep-plan` | 연구 검증 기반 계획 수립 (research → plan → verify) |
+| `/pre-generation-arch-check` | 구현 전 아키텍처 위험 점검 |
 | `/omcodex:sauron-watch` | 전체 R017 검증 |
 | `/structured-dev-cycle` | 6단계 구조적 개발 사이클 (Plan → Verify → Implement → Verify → Compound → Done) |
 | `/omcodex:lists` | 모든 사용 가능한 커맨드 표시 |
@@ -131,7 +132,7 @@ project/
 +-- AGENTS.md                    # 진입점
 +-- .codex/
 |   +-- agents/                  # 서브에이전트 정의 (48 파일)
-|   +-- skills/                  # 스킬 (108 디렉토리)
+|   +-- skills/                  # 스킬 (109 디렉토리)
 |   +-- rules/                   # 전역 규칙 (22 파일)
 |   +-- hooks/                   # 훅 스크립트 (보안, 검증, HUD)
 |   +-- contexts/                # 컨텍스트 파일 (4 파일)

package/templates/guides/browser-automation/README.md

@@ -0,0 +1,113 @@
+# Browser Automation
+
+This guide captures the browser-automation patterns that are useful in `oh-my-customcodex` without importing an external browser-control stack wholesale. It is intentionally pragmatic: keep the browser session real enough to validate user flows, but do not turn automation into a second product inside the harness.
+
+## Core Principles
+
+### 1. Prefer existing browser surfaces
+
+Start from the browser tooling the repository already has:
+
+- Playwright tests under `packages/serve/tests/`
+- the repository Playwright config at `packages/serve/playwright.config.ts`
+- the `playwright` skill and any existing MCP/browser surfaces
+
+Do not add a separate browser orchestration system unless the current surfaces are clearly insufficient.
+
+### 2. Treat authenticated sessions as sensitive
+
+Cookies, session storage, and browser profiles are credentials.
+
+- Never commit cookies, storage state, or exported browser profiles.
+- Keep imported cookies in ignored local files or ephemeral temp directories.
+- Strip secrets from screenshots, logs, and copied traces before sharing them.
+
+### 3. Anti-bot measures are for resilience, not abuse
+
+Use browser automation to validate real product flows and reproduce bugs. Do not optimize for adversarial scraping or abusive evasion.
+
+Allowed patterns:
+
+- realistic viewport and locale settings
+- waiting for stable DOM state instead of brittle sleeps
+- authenticated session reuse for testing flows behind login
+- evidence capture for failures
+
+Avoid:
+
+- aggressive stealth plugins with unclear behavior
+- retry storms against rate-limited or protected sites
+- automation that bypasses product safeguards
+
+## Cookie And Session Handling
+
+### Importing cookies
+
+When a flow requires an existing session:
+
+1. Export only the minimum cookies needed.
+2. Store them outside tracked files.
+3. Load them into a fresh browser context rather than a shared global context.
+
+### Isolate contexts
+
+Use one browser context per scenario or per worker:
+
+- prevents cross-test leakage
+- makes failures reproducible
+- keeps captured evidence attributable to one flow
+
+### Prefer storage state snapshots for repeatable tests
+
+If a workflow is run often, promote manual cookie import into a reproducible storage-state fixture rather than redoing browser login ad hoc.
+
+## Evidence Capture
+
+Browser automation is only useful if it leaves evidence behind.
+
+Capture at least one of:
+
+- screenshot on failure
+- console logs
+- network failures
+- trace or HAR when the flow is flaky
+- the exact page URL and visible state when an assertion fails
+
+When summarizing evidence for the model, preserve reference tokens and URLs so follow-up steps can still target the right page elements.
+
+## Design And Strategy Workflows
+
+### Product strategy sessions
+
+Browser automation is helpful when a product-strategy review needs to ground abstract questions in the actual product surface:
+
+- what the first-time user sees
+- where onboarding friction appears
+- how many decisions a user must make before value
+
+### Design shotgun sessions
+
+Use browser capture to compare alternatives consistently:
+
+- same viewport
+- same user state
+- same content seed
+- same evidence format
+
+That keeps visual comparisons honest and makes "taste memory" reusable.
+
+## Operational Checklist
+
+- Use an isolated browser context.
+- Keep credentials out of tracked files.
+- Capture evidence for every meaningful failure.
+- Reuse existing Playwright surfaces before adding new tooling.
+- Prefer stable selectors and deterministic waits over timing hacks.
+
+## Related Surfaces
+
+- `.codex/skills/product-strategy/SKILL.md`
+- `.codex/skills/design-shotgun/SKILL.md`
+- `.codex/skills/playwright-compress/SKILL.md`
+- `guides/web-scraping/README.md`
+- `packages/serve/playwright.config.ts`

package/templates/guides/browser-automation/index.yaml

@@ -0,0 +1,20 @@
+# Browser Automation Guide
+
+metadata:
+  name: browser-automation
+  description: Browser automation patterns for authenticated sessions, cookie handling, and evidence capture
+
+source:
+  type: internal
+
+topics:
+  - authenticated-sessions
+  - cookie-import
+  - anti-bot-caution
+  - evidence-capture
+  - session-isolation
+
+used_by:
+  - product-strategy
+  - design-shotgun
+  - playwright-compress

package/templates/guides/claude-code/14-token-efficiency.md

@@ -1,6 +1,6 @@
 # Token Efficiency Layers
 
-Token efficiency in oh-my-customcodex has three layers. They solve different problems and should not be mixed together blindly.
+Token efficiency in oh-my-customcodex has four layers. They solve different problems and should not be mixed together blindly.
 
 ## Layer 1: Plugin-Level Protection
 
@@ -22,7 +22,19 @@ Use R013 ecomode and existing runtime guards when you want to compress active-se
 
 This layer changes how the session behaves while work is running.
 
-## Layer 3: Settings-Level Optimization
+## Layer 3: Tool-Specific Compression
+
+Use `playwright-compress` when the problem is not the whole session, but one extremely verbose browser tool result.
+
+This layer is intentionally narrow:
+
+- compress verbose Playwright MCP output after the tool succeeds
+- preserve `ref=` tokens and URLs for follow-up interaction
+- keep browser evidence actionable without keeping the full raw payload in context
+
+This layer complements runtime compression instead of replacing it.
+
+## Layer 4: Settings-Level Optimization
 
 Use `/token-efficiency-audit` when you want configuration-level changes before the session burns tokens.
 
@@ -52,12 +64,14 @@ Typical settings-level levers:
 |------|------------|
 | Protect cache value across pauses | Layer 1 |
 | Compress runtime behavior in large sessions | Layer 2 |
-| Reduce baseline token spend from configuration | Layer 3 |
+| Compress one noisy browser interaction | Layer 3 |
+| Reduce baseline token spend from configuration | Layer 4 |
 
 Use layers together, but keep responsibilities separate:
 
 - plugin layer for cache and resume
 - runtime layer for in-session behavior
+- tool-specific layer for noisy browser interactions
 - settings layer for pre-session defaults
 
 ## Tradeoffs

package/templates/guides/index.yaml

@@ -34,6 +34,12 @@ guides:
       origin: github
       url: https://github.com/vercel-labs/agent-skills
 
+  - name: browser-automation
+    description: Browser automation patterns for authenticated sessions, cookie handling, and evidence capture
+    path: ./browser-automation/
+    source:
+      type: internal
+
   # Languages
   - name: golang
     description: Go language reference from Effective Go

package/templates/manifest.json

@@ -1,6 +1,6 @@
 {
-  "version": "0.1.8",
-  "lastUpdated": "2026-04-19T08:34:00.000Z",
+  "version": "0.2.0",
+  "lastUpdated": "2026-04-19T08:50:00.000Z",
   "components": [
     {
       "name": "rules",
@@ -18,13 +18,13 @@
       "name": "skills",
       "path": ".codex/skills",
       "description": "Reusable skill modules (includes slash commands)",
-      "files": 108
+      "files": 112
     },
     {
       "name": "guides",
       "path": "guides",
       "description": "Reference documentation",
-      "files": 38
+      "files": 39
     },
     {
       "name": "hooks",