Skip to content

README.md

Latest commit

 

History

History
208 lines (126 loc) · 11.3 KB

README.md

File metadata and controls

208 lines (126 loc) · 11.3 KB

Staff Engineer Mode

Release

Your AI coding agent ships fast. This makes it ship with judgment.

Staff-level engineering reasoning for design, development, testing, release, and operations decisions.

Why Now

AI coding agents now write material amounts of production code. The bottleneck is no longer how fast the agent writes — it is whether the agent reasoned about what happens when the code runs at 3am. Agents will happily design an endpoint without a compatibility plan, implement a migration with no rollback, add tests that miss the failure mode, or ship a config change with no canary. This pack closes that gap.

How It Works

Ask a normal engineering question. Hand the agent a task, design, diff, incident, rollout, or maintenance problem. The router reads the work, picks one specialist (occasionally one secondary), reads that specialist file, and returns concrete decisions, risks, checks, owners, supporting details, and next steps. You never name a specialist.

Supported tools should list only the native staff-engineer-mode router. Specialist files live under specialists/ and load only after routing.

The router refuses to load every plausible specialist. One primary specialist at a time, by default.

See SAMPLE-PROMPTS.md for prompts across every specialist.

What It Looks Like

These are real Claude Code captures from public codebases. The prompt is plain developer wording; Staff Engineer Mode routes to one specialist, loads that specialist file, and returns a structured engineering answer.

API compatibility review Claude Code capture showing an API compatibility review for an account API response change
Backup and recovery review Claude Code capture showing a backup and recovery review for a database restore plan
Code readability for agents review Claude Code capture showing a code readability for agents review for repo areas an AI coding agent may misunderstand
Dependency and code hygiene plan Claude Code capture showing a dependency and code hygiene plan for stale dependencies and dead helper code
Dependency resilience review Claude Code capture showing a dependency resilience review for remote media fetch timeouts, retries, and fallback behavior
High availability design Claude Code capture showing a high availability design review for a multi-region control plane
Observability and alerting review Claude Code capture showing an observability and alerting review for API server request errors and latency
Performance and capacity review Claude Code capture showing a performance and capacity review for an account lookup endpoint
Production readiness before launch Claude Code capture showing a production readiness review for a media processing queue split

Installation

Claude Code

Register the marketplace:

/plugin marketplace add https://github.com/sirmarkz/staff-engineer-mode.git

Install the plugin:

/plugin install staff-engineer-mode@staff-engineer-mode

Codex

Works with Codex CLI and Codex App. Tell Codex:

Fetch and follow instructions from https://raw.githubusercontent.com/sirmarkz/staff-engineer-mode/main/.codex/INSTALL.md

Cursor

/add-plugin staff-engineer-mode

OpenCode

Works with OpenCode. Tell OpenCode:

Fetch and follow instructions from https://raw.githubusercontent.com/sirmarkz/staff-engineer-mode/main/.opencode/INSTALL.md

GitHub Copilot CLI

Register the marketplace:

copilot plugin marketplace add https://github.com/sirmarkz/staff-engineer-mode.git

Install the plugin:

copilot plugin install staff-engineer-mode@staff-engineer-mode

Gemini CLI

gemini extensions install https://github.com/sirmarkz/staff-engineer-mode

Verify

Start a fresh session inside any open repo and ask one of:

  • "Before implementing partner webhooks, design the event contract, delivery retries, replay path, and dead-letter handling."
  • "During development of the checkout inventory call, decide timeout, retry, fallback, and duplicate-work safeguards."
  • "Review my last commit and tell me what you would catch in PR review."

The agent should load the router, choose one specialist, and respond with concrete decisions, risks, checks, owners, supporting details, and next steps — not vibes.

What's Inside

One native router skill: staff-engineer-mode. It routes to 54 specialist files under specialists/; those files are not installed or listed as separate native skills.

Examples by surface (the full catalog with prompts for every specialist file is in SAMPLE-PROMPTS.md):

Surface Example specialist files
Architecture and interfaces architecture-decisions, api-design-and-compatibility, data-contracts, state-machine-correctness
Reliability and resilience slo-and-error-budgets, high-availability-design, dependency-resilience, backup-and-recovery, resilience-experiments, performance-and-capacity
Delivery and change safety progressive-delivery, feature-flag-lifecycle, release-build-reproducibility, testing-and-quality-gates, test-data-engineering, dev-environment-parity, migration-and-deprecation, code-readability-for-agents, dependency-and-code-hygiene, configuration-and-automation-safety, fleet-upgrades
Operations and observability observability-and-alerting
Security and privacy secure-sdlc-and-threat-modeling, identity-and-secrets, cryptography-and-key-lifecycle, software-supply-chain-security, vulnerability-management, tenant-isolation, privacy-and-data-lifecycle
Data and workflow systems distributed-data-and-consistency, database-operations, event-workflows, data-pipeline-reliability, caching-and-derived-data
Platform and edge infrastructure-and-policy-as-code, internal-service-networking, edge-traffic-and-ddos-defense, cost-aware-reliability
Client, ML/AI, and experimentation web-release-gates, mobile-release-engineering, accessibility-gates, llm-application-security, llm-evaluation, llm-serving-cost-and-latency, ml-reliability-and-evaluation, experimentation-and-metric-guardrails
Engineering workflow, readiness, and controls agent-pr-review, ai-coding-governance, documentation-lifecycle, engineering-control-evidence, production-readiness-review, incident-response-and-postmortems, oncall-health, platform-golden-paths

Every specialist file appears in SAMPLE-PROMPTS.md with four representative prompts.

Compared To Alternatives

Staff Engineer Mode can be used alongside workflow skills like Superpowers. Workflow skills shape how the agent plans and executes; Staff Engineer Mode shapes the engineering checks and decisions it applies to the work.

Contributing

Patches welcome — especially additional practices from authoritative sources: first-party engineering publications, official documentation, standards bodies, peer-reviewed papers, or widely cited practitioner references.

New specialist files must be technology-agnostic, cite stable source IDs, and avoid vendor endorsement. Read STYLE.md before opening a PR. The voice is enforced.

Maintainers

See MAINTAINERS.md.

Sources And Influences

This pack focuses on the intersection of the strongest publicly documented engineering practices from leading software engineering organizations. It synthesizes large-operator engineering writing (Google, Amazon, Meta, Microsoft, Apple, Netflix) and standards work cited by their teams (NIST, CISA, OWASP, OpenSSF, IETF, W3C). Specific source IDs are in skills/_shared/references/source-index.md. This is an independent project; nothing here is endorsed by or affiliated with those organizations.

License

MIT — see LICENSE. The project notice is included there.

Fewer vibes. More engineering.