The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process third-party instruction files (Agent Skills) to perform audits and optimizations. This creates a surface where malicious instructions in a processed skill could attempt to influence the agent's maintenance actions, such as synchronization or versioning logic.
Ingestion points: scripts/lint_skill.py and scripts/inspect_layout.py read the full content of target SKILL.md and related files.
Boundary markers: Absent. The tools process raw file content without sandboxing or specific delimiters to isolate potential instructions from data.
Capability inventory: The skill suite has significant local capabilities: scripts/sync_installation.py performs file writes and deletions (shutil.copy2, path.unlink); scripts/bump_version.py overwrites markdown and YAML files; scripts/inspect_layout.py executes git commands.
Sanitization: The tool relies on heuristic regex and string matching which can be bypassed by sophisticated adversarial content.
[COMMAND_EXECUTION]: scripts/inspect_layout.py executes git commands via subprocess.run to inspect repository state (status, branch, toplevel). These calls are performed using list-based arguments without shell=True, reducing injection risk, but they remain a functional capability of the skill.

lov-skill-optimizer