Audit Paths

Parse the argument to determine which mode to run:

• all (or no argument) — scan memory/config files AND LaunchAgent plists
• memory — only scan CLAUDE.md, MEMORY.md, and memory/*.md
• launchagents — only scan ~/Library/LaunchAgents/*.plist

Execution

bash ~/.claude/skills/audit-paths/scripts/audit_paths.sh <mode>

Presenting Results

After running the script, present results as a table:

For each dead path found, suggest a specific fix:

• If the path looks like a renamed file/dir, suggest the likely new path
• If the path looks deleted, suggest removing the reference
• If it's a LaunchAgent, suggest updating the plist or unloading the agent

If no dead paths are found, confirm all paths are valid.

Troubleshooting

• Script exits with non-zero code: Exit code = number of dead paths found. This is expected — not an error.
• False positives for relative paths: The script resolves relative paths against the vault root. If a path is relative to a different base, it may false-positive.
• What it skips on purpose: lines containing a removal word (deleted, removed, uninstalled, retired, dropped, gone, stale, absent, left), because memory records name absent paths deliberately. It also skips globs and braces, <placeholders>, ..., $(…), /Volumes/ paths, and command strings whose first word is an existing executable. To mark a deliberately absent path, put one of those words on its line. Test changes with AUDIT_MEMORY_DIR=<fixture dir> bash scripts/audit_paths.sh memory.
• LaunchAgent paths that look dead but work: a .log path counts as alive when its folder exists, since launchd creates it on first output. Other runtime paths (PID files) may still flag; verify before fixing.
• Permission denied on LaunchAgents: iCloud-managed plists may have restricted permissions. Run with standard user permissions — don't sudo.