The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script scripts/sweep-ledger.py using uv run. This script uses subprocess.run to invoke git commands (such as git status, git ls-files, and git rev-parse) to manage a ledger of audited files. This behavior is consistent with the skill's primary purpose as a repository sweep tool.\n- [INDIRECT_PROMPT_INJECTION]: The skill audits entire repositories for errors and bugs, which requires the agent to ingest and process large amounts of untrusted data from the codebase. The instructions do not specify the use of boundary markers or sanitization techniques when handling this content, creating an attack surface where instructions embedded in the repository code could attempt to manipulate the agent's logic during the sweep.\n
Ingestion points: Files within the Git repository scope are read and inspected by the agent.\n
Boundary markers: None identified in the skill instructions.\n
Capability inventory: The skill can execute local scripts, perform Git operations (commits), and run repository-specific build and test tools.\n
Sanitization: No explicit sanitization or filtering is applied to the repository content before processing.\n- [TIME_DELAYED_CONDITIONAL]: The skill includes an 'Overnight Autonomy' feature that changes its operational mode based on the environment's local time. If the skill is started between 22:00 and 08:00, it enters an autonomous mode that suppresses user questions and approvals. While intended for long-running background tasks, this mechanism relies on a time-based condition to reduce human oversight.

fresh-eyes-sweep