The Agent Skills Directory

[COMMAND_EXECUTION]: Contains instructions for system-level CPU tuning (governor settings, turbo boost) using sudo on self-hosted runners to ensure benchmark reproducibility. These are high-privilege operations accompanied by prominent safety warnings.
Evidence: echo performance | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor in references/ci-regression.md.
[COMMAND_EXECUTION]: Recommends the cob utility for regression gating, which uses git reset internally and is documented with a warning regarding data loss.
Evidence: Mention of git reset behavior in references/ci-regression.md.
[EXTERNAL_DOWNLOADS]: Recommends the installation of several community-standard third-party Go tools for performance tracking and visualization.
Evidence: Usage of go install for benchdiff, cob, gobenchdata, and fgprof across multiple reference files.
[DATA_EXFILTRATION]: Outlines workflows involving the capture of profiling data from remote services via curl and the publication of metrics to external dashboard hosting (GitHub Pages).
Evidence: Workflow descriptions in references/pprof.md and references/ci-regression.md.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it is designed to ingest and analyze output from benchmark runs and external profiling tools.
Ingestion points: Benchmark stdout/text results and Go profile/trace binary data as described in SKILL.md and references/trace.md.
Boundary markers (absent): Processed data is not encapsulated in specific delimiters or safety wrappers before analysis.
Capability inventory: Access to the full Go toolchain, git, curl, and various performance measurement CLIs via Bash.
Sanitization (absent): No specific sanitization or instructions to ignore instructions embedded within the analyzed data are provided.

backend-go-benchmark