The Agent Skills Directory

[SAFE]: The skill defines a robust security model that strictly separates inspection from mutation and publication authority. It explicitly prohibits unauthorized network access, data exfiltration, or the relaxation of security controls during performance tuning.
[EXTERNAL_DOWNLOADS]: The skill references external repositories and tools for benchmarking and profiling, such as Cloudflare's Hyperfine, standard language runtimes (Zig, Rust, Go, Python, Node.js), and vendor-specific tools (e.g., 'tkersey/tap/lift'). All external references target trusted organizations, well-known services, or the skill author's own verified infrastructure.
[COMMAND_EXECUTION]: The instructions include templates for executing common profiling utilities (e.g., perf, strace, heaptrack) and language-specific test harnesses. These executions are legitimate, scoped to the skill's primary purpose of performance measurement, and include warnings against unauthorized tool installation or privileged execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external evidence like logs, traces, and task outcomes. To mitigate risks associated with this attack surface, the instructions mandate the redaction of sensitive data and the preservation of tenant isolation and data privacy boundaries.

tune