support: simplify version.sh — patch/minor/major auto-bump

- patch/minor/major: auto-detect version, bump, tag, update env/pins
- No manual version entry needed
- Uses sed instead of grep -P for portability
- Validates repo state (branches, git tracking)
- Updated .gitignore for workspace/ (ignored, recreatable)

Author: cfxdevkit

.env

@@ -0,0 +1,53 @@
+# LocalPibox Devstack — environment variables
+# All variables use the LPB_ (LocalPibox) prefix to avoid conflicts with
+# user environment variables. Loaded by start.sh from workspace directory.
+
+# ─── API Keys ────────────────────────────────────────────────────────────────
+
+# Exa MCP search API key
+LPB_EXA_API_KEY=57cd21c4-35e6-4965-8b03-49c2c56416de
+
+# ─── Editor ──────────────────────────────────────────────────────────────────
+
+# OpenVSCode connection token
+LPB_CONNECTION_TOKEN=rvKp8xMq2vNd7Jc4WzFg9Ht3Yb5EsL6A
+
+# Editor host — 0.0.0.0 = LAN access, 127.0.0.1 = local-only
+LPB_EDITOR_HOST=0.0.0.0
+
+# Editor port
+LPB_ED_PORT=3000
+
+# ─── Browser ─────────────────────────────────────────────────────────────────
+
+# Agent-browser session — uses PI_WORKTREE_ID for worktree isolation
+LPB_AGENT_BROWSER_SESSION=${PI_WORKTREE_ID}
+
+# Chrome launch args (container-safe defaults)
+LPB_AGENT_BROWSER_ARGS="--no-sandbox,--no-first-run,--disable-gpu,--disable-crashpad"
+
+# Max output length to prevent context flooding
+LPB_AGENT_BROWSER_MAX_OUTPUT=4000
+
+# LLM safety markers in content extraction
+LPB_AGENT_BROWSER_CONTENT_BOUNDARIES=true
+
+# Action categories requiring approval
+LPB_AGENT_BROWSER_CONFIRM_ACTIONS="delete,download,cookie_delete,file_access"
+
+# Daemon idle timeout (5 minutes) — auto-cleanup after inactivity
+LPB_AGENT_BROWSER_IDLE_TIMEOUT_MS=300000
+
+# ─── Context window / max tokens ratio ──────────────────────────────────────
+# max_tokens = context_window × LPB_MAX_TOKENS_CONTEXT_RATIO
+# 0.06 (6%) — reduced from 0.125 for Qwen thinking models.
+# At 0.125, Qwen thinking models generate 10-20k-token thinking blocks
+# that push prompt+max_tokens past the 262k window limit.
+# 0.06 × 262k = ~15.7k maxTokens — leaves room for thinking within window.
+LPB_MAX_TOKENS_CONTEXT_RATIO=0.06
+
+# ─── Persistence ─────────────────────────────────────────────────────────────
+
+# Persist gh CLI auth to ~/.config/gh inside container, stored in
+# ~/.localpibox/state/gh-config on the host (in the state volume).
+LPB_PERSIST_GH_CONFIG=true

.pi/mcp.json

@@ -1 +1 @@
-0.2.0-lpb
+1.0.0-lpb

PRODUCTION-VALIDATION-CHECKLIST.md

@@ -14,7 +14,7 @@
 # Fork URL — cumulative patches on top of earendil-works/pi
 LPB_PI_FORK=https://github.com/localpibox/pi.git
 # Branch — which branch to build from (the lpb branch is the stable base)
-LPB_PI_REF=lpb
+LPB_PI_REF=1.0.0-lpb
 # Upstream — original repo (for reference / rebasing)
 LPB_PI_UPSTREAM=https://github.com/earendil-works/pi.git