The Agent Skills Directory

[SAFE]: The skill serves its stated purpose as a professional engineering guide. It contains legitimate Python code implementations for automotive planning algorithms, such as Frenet-frame lattice planners and Model Predictive Control. No malicious patterns, such as prompt injections, data exfiltration, or unauthorized network access, were detected.- [INDIRECT_PROMPT_INJECTION]: The skill defines ingestion points for untrusted data in references/scenarios.md (via agent_observation in ContingencyPlanner) and SKILL.md (via integration with perception object lists). While boundary markers are absent, the capability inventory is restricted to local mathematical computation and trajectory generation across all files, and the logic performs numerical validation on inputs, effectively mitigating injection risks. No specific sanitization logic is present, but inputs are treated as float values for Bayesian updates.- [COMMAND_EXECUTION]: Instructions in SKILL.md describe modifying a local configuration file (CLAUDE.md) to establish the agent's persona. This is a documented platform feature for agent persistence and does not represent an unauthorized command execution risk.

planning-decision-engineer