The Agent Skills Directory

[DATA_EXPOSURE]: The skill accesses the local Atuin history database (history.db). Shell history often contains sensitive data such as API keys, passwords, or environment variables accidentally typed into the terminal.
[COMMAND_EXECUTION]: The script atuin_history_cleanup.py uses subprocess.run to execute various Atuin CLI commands, including atuin info, atuin store push, and atuin sync. These commands facilitate data synchronization with Atuin's remote servers.
[COMMAND_EXECUTION]: The transactional helper script atuin_history_cleanup_transactional.py utilizes a pseudo-terminal (pty) to programmatically drive the interactive Atuin TUI inspector, performing precise deletions by simulating keyboard inputs (e.g., Ctrl+O, Ctrl+D).
[DYNAMIC_EXECUTION]: The primary script uses importlib.util to dynamically load its sibling helper script atuin_history_cleanup_transactional.py at runtime. While the file is local to the skill, this pattern allows for executable code to be loaded from a path determined during execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (shell history) to generate cleanup suggestions.
Ingestion points: The load_history_entries function in scripts/atuin_history_cleanup.py reads row data from the SQLite database.
Boundary markers: Audit reports are provided in structured text or JSON formats, though they do not include explicit instructions for the LLM to ignore embedded content within the history records themselves.
Capability inventory: The skill has the ability to execute arbitrary shell commands via subprocess.run and manage database state.
Sanitization: The script uses shlex.quote for shell safety and applies conservative typo-detection heuristics (e.g., is_high_confidence_typo_token) to minimize the risk of malicious commands being interpreted as legitimate cleanup candidates.

cleaning-atuin-history