The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute a local Python script (claude_account_migrate.py) to perform its primary function of scanning and copying session files.
[INDIRECT_PROMPT_INJECTION]: The skill processes session metadata (titles) and conversation transcripts stored in the Claude desktop data directory. This creates an attack surface where malicious instructions embedded in session data could potentially influence the agent's behavior during the migration process.
Ingestion points: The script reads local_*.json metadata files and conversation transcripts from the projects/ subdirectory within the Claude configuration folder.
Boundary markers: The skill instructions define structured steps for the agent but do not implement explicit data delimiters or instructions to ignore content within the migrated session data.
Capability inventory: The migration script utilizes shutil.copy2, shutil.copytree, os.replace, and os.mkdir to manage files.
Sanitization: No specific filtering or escaping is applied to session titles or transcript content before they are presented to the agent in the migration plan.
[DATA_EXPOSURE]: The skill accesses sensitive application paths, including ~/Library/Application Support/Claude and ~/.claude. However, this access is strictly limited to the skill's stated purpose of session migration. The implementation includes specific logic to avoid credentials and tokens, and the script contains no network capabilities, ensuring all data remains local.

lov-claude-account-migrate