engineering-culture（工程文化）

Engineering Culture

Scope

Covers

• Diagnosing the current engineering culture *and* delivery system (technical, architectural, cultural, and management capabilities)
• Defining a clear engineering culture code (principles → behaviors → decision rules → anti-patterns)
• Aligning org structure with architecture (Conway’s Law) and reducing cross-team friction
• Increasing clock speed (safe shipping + experimentation throughput) and improving DevEx
• Creating a practical cross-functional workflow contract (how engineering + PM/Design/Marketing collaborate in the same toolchain)
• Making AI-assisted development safe and effective (humans as “architects”: spec, review, and oversight)

When to use

• “Help me improve engineering culture / DevEx and make it concrete.”
• “Our delivery is slow—build a plan to increase shipping speed without breaking things.”
• “Our org structure fights our architecture—analyze Conway’s Law and propose changes.”
• “We want tighter processes and faster experimentation (higher clock speed).”
• “Non-engineering functions struggle to work with engineering—define a shared workflow contract.”
• “We’re adopting AI coding tools/agents—set norms so engineers shift toward higher-level design and review.”

When NOT to use

• You need to respond to an active incident or outage (use incident response/runbooks)
• You need HR/legal policy, investigations, or employee relations handling (involve HR/legal)
• You only need to implement a specific technical improvement (e.g., “set up CI”) without culture/org/process work
• You need a full company strategy/roadmap prioritization across many bets (use prioritizing-roadmap)

Inputs

Minimum required

• Org context: product(s), stage, engineering size, team topology, on-call model
• Current symptoms + 2–5 examples (e.g., slow delivery, flaky deploys, low ownership, poor collaboration, high toil)
• Current delivery system snapshot (release cadence, CI/CD maturity, test strategy, environments)
• Architecture constraints (e.g., monolith vs services; coupling hotspots; ownership boundaries)
• Cross-functional workflow reality (where work is tracked, how decisions are made, how releases happen)
• Desired outcomes (what should be *more true* in 4–12 weeks?) + timeline constraints

Missing-info strategy

• Ask up to 5 questions from references/INTAKE.md (3–5 at a time), then proceed with explicit assumptions.
• If metrics are missing, use best-effort ranges and label confidence; list instrumentation gaps.
• Do not request secrets, credentials, or proprietary identifiers; use redacted summaries.

Outputs (deliverables)

Produce an Engineering Culture Operating System Pack in Markdown (in-chat; or as files if requested):

1. Culture + capability snapshot (what’s true today; evidence; capability gaps)
2. Engineering culture code (v1) (3–7 principles with behaviors, do/don’t, decision rules, anti-patterns)
3. Org ↔ architecture alignment brief (Conway’s Law analysis + proposed operating model changes)
4. Clock speed + DevEx improvement backlog (prioritized initiatives with owners, sequencing, metrics)
5. Cross-functional workflow contract (GitHub/issue/PR/release norms; how non-engineers contribute; AI norms)
6. Rollout + measurement plan (30/60/90, rituals, metrics + guardrails, feedback loops)
7. Risks / Open questions / Next steps (always included)

Templates: references/TEMPLATES.md Expanded guidance: references/WORKFLOW.md

Workflow (7 steps)

1) Intake + boundary setting

• Inputs: User context; references/INTAKE.md.
• Actions: Confirm scope (team vs org), decision owner(s), timeline, and constraints. Identify any HR/legal or active-incident concerns and route appropriately. Confirm which deliverables to produce.
• Outputs: Context snapshot + assumptions/unknowns list.
• Checks: Scope boundaries are explicit; success definition is stated in observable terms.

2) Diagnose the current culture as a delivery system (capability map)

• Inputs: Symptoms/examples; current process/tooling; architecture context.
• Actions: Build a capability map across technical, architectural, cultural, and management capabilities. Capture evidence and gaps (not platitudes). Distinguish stated culture vs lived culture.
• Outputs: Culture + capability snapshot (draft).
• Checks: Each claimed problem has at least one piece of evidence (example, metric, observed behavior) or is labeled “needs data”.

3) Define the target culture (culture code v1)

• Inputs: Snapshot; constraints; what already works.
• Actions: Pick 2–4 priority shifts, then write a culture code: 3–7 principles with behaviors, do/don’t, decision rules, and anti-patterns. Prefer rules that increase autonomy while reducing ambiguity.
• Outputs: Engineering culture code (v1).
• Checks: Every principle includes a concrete “how we work” example and at least one measurable/observable signal.

4) Align org structure with architecture (Conway’s Law)

• Inputs: Current team topology; architecture coupling/ownership hotspots; dependency pain.
• Actions: Map org → architecture fit. Propose changes: team boundaries, ownership, interfaces, and standardization (e.g., leveling definitions, incident policies, review expectations) where misalignment causes friction.
• Outputs: Org ↔ architecture alignment brief.
• Checks: Proposed changes include migration/transition steps and explicit trade-offs (what gets worse).

5) Increase clock speed (safe shipping + experimentation throughput)

• Inputs: Current shipping/experiment cadence; pipeline constraints; quality constraints.
• Actions: Define “clock speed” targets and bottlenecks. Propose initiatives that raise throughput safely (small batches, CI reliability, test strategy, progressive delivery, observability). Convert into a prioritized backlog.
• Outputs: Clock speed + DevEx improvement backlog (draft).
• Checks: Each initiative has an owner, an effort range, a dependency note, and a metric/leading indicator.

6) Create the workflow contract (including AI norms)

• Inputs: Collaboration pain points; tool constraints; roles.
• Actions: Specify how work flows from idea → issue → PR → deploy → learn. Define cross-functional participation (where PM/Design/Marketing contribute) and working agreements (review SLAs, merge/deploy policy, experiment ownership). Add AI-assisted development norms: where agents help, human review requirements, and safe data handling.
• Outputs: Cross-functional workflow contract.
• Checks: The contract reduces common failure modes (stalled PRs, unclear ownership, “drive-by” requests) and is teachable to new hires.

7) Rollout + measurement + quality gate

• Inputs: Draft pack.
• Actions: Create a 30/60/90 rollout plan with rituals/cadence and training. Define metrics and guardrails (e.g., DORA + quality + DevEx). Run references/CHECKLISTS.md and score with references/RUBRIC.md. Finalize Risks / Open questions / Next steps.
• Outputs: Final Engineering Culture Operating System Pack.
• Checks: The first 1–2 actions can start this week; measurement is feasible; risks/trade-offs are explicit.

Quality gate (required)

• Use references/CHECKLISTS.md and references/RUBRIC.md.
• Always include: Risks, Open questions, Next steps.

Examples

Example 1 (slow delivery + DevEx): “Use engineering-culture. Context: B2B SaaS, 35 engineers, monolith + a few services, weekly releases, rising incidents. Goal: increase shipping speed without quality regressions. Output: an Engineering Culture Operating System Pack with a clock-speed backlog and a workflow contract.”

Example 2 (Conway misalignment): “We have 6 teams but architecture ownership is unclear and everything depends on platform. Analyze Conway’s Law issues and propose a new operating model + standardization (leveling, code ownership, on-call) plus a rollout plan.”

Boundary example: “Write a generic essay about what engineering culture is.” Response: explain this skill produces a concrete operating system pack; ask for context/symptoms/timeline or provide the intake checklist and an example template from references/TEMPLATES.md.