golang-benchmark

Pass

Audited by Gen Agent Trust Hub on Apr 29, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides detailed instructions for executing Golang performance measurement commands, including running benchmarks (go test -bench), generating profiles (go tool pprof, go tool trace), and performing statistical comparisons (benchstat). It also includes specific commands for system-level tuning on dedicated CI runners (e.g., setting CPU governors and disabling Turbo Boost) using sudo, which are accompanied by clear safety warnings and intended for specialized environments.
  • [EXTERNAL_DOWNLOADS]: Directs the user to install several reputable Golang performance utilities from external repositories. These include benchstat from golang.org (Google-managed), benchdiff from filippo.io (Filippo Valsorda), cob from github.com/knqyf263, and gobenchdata from go.bobheadxi.dev. These sources are recognized as standard within the Go development community.
  • [DATA_EXFILTRATION]: Mentions the use of curl to capture diagnostic trace and profile data from local endpoints (e.g., localhost:6060/debug/pprof/trace). This is a standard procedure for local application diagnostics and does not involve unauthorized remote data transmission.
  • [PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection by instructing the agent to process data from external sources.
  • Ingestion points: Reads and analyzes benchmark results from local files (e.g., old.txt, new.txt, bench.txt) as described in SKILL.md and references/benchstat.md.
  • Boundary markers: Absent; the instructions do not specify the use of delimiters or 'ignore' commands when processing benchmark output.
  • Capability inventory: The skill possesses the capability to execute a variety of command-line tools via Bash across multiple scripts.
  • Sanitization: Absent; benchmark results are interpreted as raw text without automated validation or sanitization routines.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 29, 2026, 03:11 PM