feat(skills): add cleanup-artifacts skill and session hygiene protocol

bin/sync-skills

@@ -8,11 +8,17 @@ import sys
 
 
 def lexical(path):
-    return Path(os.path.abspath(path))
+    s = str(os.path.abspath(path))
+    if s.startswith('\\\\?\\'):
+        s = s[4:]
+    return Path(s)
 
 
 def link_target(path):
-    return lexical(path.parent / os.readlink(path))
+    target = os.readlink(path)
+    if os.path.isabs(target):
+        return lexical(target)
+    return lexical(path.parent / target)
 
 
 def inside(path, root):

bin/validate-skills

@@ -51,7 +51,7 @@ def metadata(text):
 
 def local_links(path):
     fenced = False
-    for number, line in enumerate(path.read_text().splitlines(), 1):
+    for number, line in enumerate(path.read_text(encoding='utf-8').splitlines(), 1):
         if line.lstrip().startswith(('```', '~~~')):
             fenced = not fenced
             continue

protocols/workflow.md

@@ -43,3 +43,10 @@ Choose checks that can detect a real failure in the changed behavior.
 
 Complete means the requested behavior is implemented and supported by empirical
 evidence. A successful command exit alone does not prove the user-facing outcome.
+
+## Session Hygiene
+
+After task verification and handoff are complete, offer or execute (upon user
+confirmation) session cleanup via `cleanup-artifacts`. Purge scratch scripts,
+stale test caches, and harness session bloat across Antigravity, Claude Code,
+Codex, and OpenCode while strictly preserving rules, custom skills, and MCP tools.

skills/cleanup-artifacts/SKILL.md

@@ -0,0 +1,43 @@
+---
+name: cleanup-artifacts
+description: >
+  Clean up temporary session files, tool caches, browser recordings, and conversation bloat across agent harnesses after user confirmation.
+scope: global
+---
+
+# Cleanup Artifacts & Session Hygiene
+
+Use this skill after task implementation and verification are complete, when the user requests or confirms cleaning up session bloat across agent harnesses (Antigravity, Claude Code, Codex, OpenCode/Aider).
+
+## Principles
+
+1. **User Confirmation Required**: Never purge conversation histories, scratch files, or caches without explicit user intent.
+2. **Preserve Rules, Agents, MCPs & Projects**: Never remove configuration files, rules, custom agents, MCP tool schemas, database migrations, or project repositories.
+3. **Preserve Active Session**: If operating inside an active agent conversation, preserve the active session ID's brain and transcript.
+
+## Targets by Harness
+
+### 1. Antigravity
+- **Past Conversations**: Purge inactive conversation databases in `%USERPROFILE%\.gemini\antigravity\conversations\`.
+- **Inactive Brains**: Purge non-active UUID folders in `%USERPROFILE%\.gemini\antigravity\brain\`.
+- **Temporary Tool & Browser Traces**: Purge `%USERPROFILE%\.gemini\antigravity\browser_recordings\*`, `cache\*`, `artifacts\*`, `implicit\*`, and `sidecar_data\*`.
+- **Installers & Staged Updates**: Purge `%LOCALAPPDATA%\Temp\antigravity-stable-user-x64` and `%LOCALAPPDATA%\antigravity-updater\*`.
+- **Web & GPU Caches (Deep Clean)**: Optionally flush `Cache`, `Code Cache`, `GPUCache`, and `CachedData` inside `%APPDATA%\Antigravity` and `%APPDATA%\Antigravity IDE`.
+
+### 2. Claude Code
+- **Caches & Transcripts**: Clear inactive sessions in `~/.claude/artifacts/`, ephemeral scratchpads, and debug log files.
+- **Preserve**: Keep `~/.claude/settings.json`, configured skills, and authenticated profiles.
+
+### 3. Codex & OpenCode / Aider
+- **Scratch & Diffs**: Purge stale patch chunks (`.patch`, `.diff.tmp`), transient tool logs, and cached completions in `.codex/cache/` or `.opencode/`.
+
+### 4. Workspace Level
+- Purge intermediate scratch test scripts (`scratch/*.ps1`, `scratch/*.py`) created solely for one-off debugging.
+- Purge standard build/test cache directories generated during task verification (e.g., `.pytest_cache`, `.coverage`, `__pycache__`, `node_modules/.cache`) when confirmed.
+
+## Workflow
+
+1. **Assess & Report**: Check current file counts and sizes in the relevant harness locations. Present the summary to the user.
+2. **Confirm Scope**: Clarify whether to run standard cleanup (session data, past artifacts, temp files) or deep cleanup (including IDE/browser GPU caches).
+3. **Execute**: Remove the target paths cleanly using non-interactive commands.
+4. **Report Outcome**: Output a before-and-after reclaimed disk space summary.