The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Python scripts (scripts/platforms.py, scripts/bookmarks.py) and standard system utilities (mkdir, cp, curl) to manage bookmark files. These commands are scoped to specific bookmark storage locations and are used for locating files, checking process status, creating backups, and applying transformations.
[PROMPT_INJECTION]: The SKILL.md and references/safe-mutation-rules.md files include specific instructions to treat bookmark titles and file content as data only, explicitly warning against 'ignore-previous-instructions' style patterns. While a static detector flagged an 'override' pattern, in context, this is a defensive instruction strengthening the agent's resistance to injection attacks.
[PRIVILEGE_ESCALATION]: The skill identifies when macOS Full Disk Access is required but does not attempt to bypass these permissions. Instead, it instructs the user on how to manually grant the necessary access via System Settings.
[DATA_EXFILTRATION]: The skill uses curl to check for dead links if requested by the user. The operation is limited to fetching HTTP headers (-I) and treats timeouts or access denials neutrally, minimizing data exposure.
[SAFE]: The skill follows a strict 'Safe Mutation Workflow' which includes: never hardcoding paths, performing pre-flight checks, creating mandatory backups before any modification, and using the system trash/recycle bin instead of permanent deletion.

organizing-bookmarks