The Agent Skills Directory

[SAFE]: The skill is a procedural guide for performance engineering. It promotes best practices such as establishing a baseline before any code changes and using identical measurement protocols to ensure data integrity.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands for benchmarking and running test suites. These commands are dynamically resolved from a local project configuration file (AGENTS.md). This is a standard functional requirement for development-oriented agents and does not involve hardcoded or remote-sourced commands.
[PROMPT_INJECTION]: The skill processes external inputs such as 'task packets' and 'change plans', which constitutes an indirect prompt injection surface.
Ingestion points: Data enters the agent context through the referenced task packets and change plans.
Boundary markers: The skill utilizes a structured markdown template (references/task-template.md) with clear headers and {{placeholder}} syntax to separate external data from instructions.
Capability inventory: The agent is authorized to execute commands defined in the local AGENTS.md file (benchmarks and suites).
Sanitization: The skill relies on numeric verification and human-in-the-loop review of output logs and code diffs rather than automated input filtering.

write-performance