Imported from rsturla/skills (AGENTS.md). Install upstream with npx skills add rsturla/skills. Copyright stays with the author.

Coding Style

General

• No trailing whitespace
• Files end with a single newline
• UTF-8 encoding everywhere
• Lines max 120 chars (code), 120 chars (prose/markdown)

Naming

• Files and directories: kebab-case (new-worktree, commit-message)
• Branches: dev/robertsturla/<change> with kebab-case change name
• Commits: Conventional Commits format, subject ≤50 chars, body lines ≤72 chars

Shell (bash)

• set -uo pipefail (avoid set -e in scripts with arithmetic or grep)
• Use [[ not [ for conditionals
• Quote all variable expansions: "$var" not $var
• Heredocs for multi-line strings, not nested quotes
• Functions before main logic
• Prefer command -v over which

Markdown (SKILL.md)

• YAML frontmatter: name, description, compatibility, allowed-tools, metadata
• One blank line between sections
• Fenced code blocks with language tag
• Tables for structured reference data
• No emoji unless explicitly requested

Git

• Conventional Commits: feat/fix/refactor/docs/test/chore/ci/build/perf/style
• Imperative mood in subject line, lowercase after prefix (feat: add thing not feat: Add thing)
• No trailing period on subject
• Body explains why, not what
• Atomic commits — one logical change per commit, each commit should build and pass tests independently
• Never amend published commits
• To fix a previous commit, use git commit --fixup=<sha> then git rebase -i --autosquash before pushing
• Prefer fixup over amend — keeps history clean while preserving the ability to review intermediate states

Language Guides

• Go Coding Guidelines
• Rust Coding Guidelines
• Python Coding Guidelines
• OpenTofu + Terragrunt Guidelines
• Containerfile Guidelines
• GitHub Actions Guidelines
• API Design Guidelines
• AWS Guidelines — when asked anything about AWS, read the relevant files in docs/aws/ first

Tool & Language Preferences

Apply only the preferences relevant to the question. Do not list unrelated preferences.

• Containers: podman, never docker. Use Containerfile, never Dockerfile
• Container registry: Quay, not ECR (ECR only when AWS service requires it)
• Kubernetes: OpenShift (platform-managed), not EKS
• Secrets: Vault (platform-managed), not AWS Secrets Manager/Parameter Store
• VCS: git
• Languages (order of preference): Go → Rust → Python
• IaC: OpenTofu + Terragrunt, never Terraform. CloudFormation acceptable
• JS/TS runtime: Bun, never npm/yarn/pnpm
• LLM APIs: Vertex AI only (GCP project itpc-gcp-core-pe-eng-claude). No direct Anthropic API, OpenAI, Copilot, or other LLM provider access. Always use the Vertex AI rawPredict endpoint with the global endpoint (aiplatform.googleapis.com, not regional). Model IDs: claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5. No date suffixes, no @date patterns — bare short names only. See the vertex-ai skill for details.
• Telemetry/Observability: OpenTelemetry only. Instrument with OTel SDK, export via OTLP. When integrating with Datadog, use OTel (DDOT Collector or Agent OTLP ingest) — not Datadog-native SDKs. Only fall back to DD SDK for features that require it (Continuous Profiler, App Security). DD OTLP metrics intake requires delta temporality.
• Shell: bash for scripts
• Package installs: NEVER run any system package manager (dnf, brew, apt, yum, pacman, zypper, etc.). If a tool is missing, ask the user to provide it.

Dependencies

• In applications: use proper language SDKs (Go, Rust, Python). Do NOT shell out to curl/jq from application code.
• In skills and git hooks: use standard unix tools (curl, jq, gcloud, oc) since skills are markdown instructions and hooks should avoid runtime dependencies.