yet-another-agent-harness: Skill for Claude Code

Agent-Ready Go

Make Go applications work effectively with AI coding agents by ensuring all tooling produces machine-readable output, errors are structured, and commands run non-interactively by default.

Workflow

Making a Go app agent-ready involves these steps:

1. Audit — identify gaps against the agent-readiness checklist
2. Structured logging — add/configure slog, Zap, ZeroLog, or Logrus with JSON output
3. Machine-readable output — ensure all commands can emit JSON
4. Linting — add .golangci.yml with thorough config
5. Testing — ensure go test -race -count=1 -timeout=5m with coverage thresholds
6. Error handling — structured errors, meaningful exit codes
7. Non-interactive design — --yes/-y flags, env vars over prompts
8. Makefile — standardize build, test, lint, cover targets

For greenfield projects, apply all steps. For existing projects, run the audit first and address only what's missing.

Step 1: Audit

Run this checklist against the project. Address each ❌:

• Structured logger configured (slog / Zap / ZeroLog / Logrus)
• Logger outputs JSON when not a TTY or when LOG_FORMAT=json
• CLI commands support --json or --output json flag
• .golangci.yml exists with a thorough linter set
• go test -race passes
• Code coverage ≥ 80% enforced
• go vet ./... passes clean
• All errors wrapped with context (fmt.Errorf("...: %w", err))
• Exit codes: 0 = success, non-zero = failure (no panics in CLI entry)
• context.Context threaded through all I/O-bound functions
• Interactive prompts have --yes/-y bypass flag
• Makefile with build, test, lint, cover, ci targets
• Graceful shutdown on SIGTERM/SIGINT (HTTP services)
• Health check endpoints (/healthz, /readyz) for HTTP services
• govulncheck ./... passes clean
• NO_COLOR respected; no ANSI codes in non-TTY output
• Config loaded from env vars with validation at startup