The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes untrusted data in the form of Roblox place files (.rbxl, .rbxlx) and Luau script source code during its performance and static auditing procedures.
Ingestion points: PerfAudit.luau reads the Source property of LuaSourceContainer objects; lune_perf_audit.luau deserializes entire place files via the Lune @lune/roblox library.
Boundary markers: The skill does not currently implement explicit boundary markers or "ignore embedded instructions" warnings when passing source code content or audit findings back to the agent context.
Capability inventory: The skill has the ability to execute shell commands via subprocess in Python and execute Luau code within a Roblox Studio session using the rbx_run toolkit.
Sanitization: While the skill performs regex-based pattern matching to find performance issues, it does not sanitize or escape the content of scripts before presenting them to the agent.
[COMMAND_EXECUTION]: The Python runner scripts/rbx_perf.py executes the lune binary to perform offline audits of Roblox place files.
Evidence: The audit_place function in scripts/rbx_perf.py uses subprocess.run to invoke the Lune CLI runtime.
[DYNAMIC_EXECUTION]: The skill dynamically assembles Luau executable code by inlining local "AgentKit" modules and concatenating them with agent-supplied code strings before execution in Roblox Studio.
Evidence: The kit_source and run functions in scripts/rbx_perf.py perform runtime assembly of Luau source code from template files.

scenario-roblox-performance