Add local skill metadata index

.gitignore

@@ -26,6 +26,9 @@ Thumbs.db
 auth.json
 model_responses.txt
 
+# Sensitive files (API keys, credentials)
+mykey.py
+
 tasks/
 
 *.zip
@@ -36,29 +39,82 @@ memory/*
 # Allow tracking of specific SOPs
 !memory/web_setup_sop.md
 !memory/autonomous_operation_sop.md
+!memory/autonomous_operation_sop/
+!memory/autonomous_operation_sop/**
 !memory/scheduled_task_sop.md
 
+# L4 session archiver (only the script, not archives)
+!memory/L4_raw_sessions/
+memory/L4_raw_sessions/*
+!memory/L4_raw_sessions/compress_session.py
+
 # ljqCtrl related tools
 !memory/ljqCtrl.py
 !memory/ljqCtrl_sop.md
 
-# mem_scanner related tools
-!memory/mem_scanner.py
-!memory/mem_scanner_sop.md
+# procmem_scanner related tools
+!memory/procmem_scanner.py
+!memory/procmem_scanner_sop.md
 
 # TMWebDriver SOP
 !memory/tmwebdriver_sop.md
 
+# Vue3 Component SOP
+!memory/vue3_component_sop.md
+
 # Subagent SOP
 !memory/subagent_sop.md
 
+# Supervisor SOP
+!memory/supervisor_sop.md
+
+# Plan SOP
+!memory/plan_sop.md
+
+# Goal Mode SOP
+!memory/goal_mode_sop.md
+
+# Skill Search SOP
+!memory/skill_search/
+!memory/skill_search/**
+
+
 # ADB UI tool
 !memory/adb_ui.py
 
+# Keychain
+!memory/keychain.py
+
+# Vision / OCR / UI detection tools
+!memory/ocr_utils.py
+!memory/vision_sop.md
+!memory/ui_detect.py
+!memory/vision_api.template.py
+
+# Memory management
+!memory/memory_cleanup_sop.md
+
+# Code Review Principles
+!memory/code_review_principles.md
+
 # Visual Studio
 .vs/
 restore_commit.txt
 
 sche_tasks/
-model_responses.sync-conflict-*
-QUICK_START.md
+# CDP Bridge 密钥配置（首次运行自动生成）
+assets/tmwd_cdp_bridge/config.js
+assets/copilot_proxy.pyw
+**log.*
+
+# Reflect (ignore new files, whitelist existing)
+reflect/*
+!reflect/autonomous.py
+!reflect/scheduler.py
+!reflect/agent_team_worker.py
+!reflect/goal_mode.py
+
+# Universal: never track __pycache__ anywhere
+**/__pycache__/
+
+.claude/

CONTRIBUTING.md

@@ -0,0 +1,44 @@
+# Contributing to GenericAgent
+
+## Why This File Is Short
+
+GenericAgent's core is ~3K lines. Every file in this repo will be read by AI agents — potentially thousands of times. Extra words cost real tokens and push useful context out of the window, increasing hallucinations. This document practices what it preaches: **say only what matters.**
+
+## Before You Contribute
+
+1. **Read the codebase first.** It's small enough to read in one sitting. Understand the philosophy before proposing changes.
+2. **Open an Issue first** for anything non-trivial. Discuss before coding.
+
+## Code Standards
+
+All PRs go through a strict automated code review skill. Key expectations:
+
+- **Self-documenting code, minimal comments.** If code needs a paragraph to explain, rewrite it.
+- **Compact and visually uniform.** Fewer lines, consistent line lengths, no fluff.
+- **Small change radius.** Changing A shouldn't ripple through B, C, D.
+- **More features → less code.** Good abstractions make the codebase shrink, not grow.
+- **Let it crash by failure radius.** Critical errors fail loud; trivial ones pass silently. No blanket try-catch.
+
+> ⚠️ This review is deliberately strict — most AI-generated code (e.g. Claude Code output) will not pass as-is. Read the full principles before submitting.
+
+## Skill Contributions
+
+GenericAgent evolves through skills. Not all skills belong in the core repo:
+
+| Type | Where it goes | Example |
+|---|---|---|
+| **Fundamental / universal** | Core repo (`memory/`) | File search, clipboard, basic web ops |
+| **Domain-specific / niche** | Skill Marketplace *(coming soon)* | Stock screening, food delivery, specific API integrations |
+
+If your skill only makes sense for a specific workflow, it's a marketplace candidate, not a core PR.
+
+## PR Checklist
+
+- [ ] Issue linked or context explained in ≤3 sentences
+- [ ] Code passes the [review principles] self-check:
+  1. Can I safely modify this locally without reading the whole codebase?
+  2. Is there a clear core abstraction — new features add implementations, not modify old logic?
+  3. Are change points converging at boundaries, not scattered everywhere?
+  4. On failure, can I quickly locate the responsible module?
+- [ ] Net line count: ideally negative or zero for refactors
+- [ ] No unnecessary dependencies added