Code discipline

Code Discipline

A coding methodology for production systems that must last. Not a communication standard. Not an honesty protocol. Purely: how to write code well.

Tradeoff: This discipline biases toward correctness and simplicity over speed and cleverness. For throwaway scripts, apply judgment. For production code — apply everything.

---

Priority Order

Every technical decision is evaluated in this order:

1. Stability
2. Reliability
3. Readability
4. Maintainability
5. Scalability
6. Performance
7. Elegance

The goal is not "it works" — it is "it works correctly and reliably one year from now."

---

1. Think Before Coding

Don't assume. Surface tradeoffs. Map blast radius.

Before writing any code, present all four:

1. Working assumptions — what are you assuming to be true?
2. Constraints and risks — what could go wrong?
3. Multiple valid solutions? — if yes, stop and present options; do not choose independently
4. Dependencies and blast radius — what else is affected by this change?

Rules:

• If uncertain — ask, don't assume and proceed
• If multiple interpretations exist — present them, don't pick silently
• If a simpler approach exists — say so and push back
• If information is missing — stop and ask; do not proceed without answers

---

2. Verify Reality Before Acting

Don't code from memory. The codebase is the source of truth.

Before writing or changing code:

• Read the actual files involved. Don't work from a remembered shape.
• Don't invent APIs, function names, type names, field names, environment variables, configuration keys, or imports.
• Don't assume a pattern, helper, or utility exists unless you verified it in the repository.
• Don't assume a third-party library behaves a certain way without checking its documentation or signature.
• If you cannot find evidence in the codebase, say so explicitly: "I couldn't find this in the code" — and stop. Do not fill the gap from memory.

The test: Every function call, import, type, and field in your code must trace to a verified source — an actual file, an actual signature, or documentation you read.

Evidence standard for claims: When stating that code works, that a test passes, or that a file does something specific, cite file:line and the quoted code or output. Paste actual command output rather than summarizing it. "It should work" and "I believe" are not evidence.

---

3. Simplicity First

Minimum code that solves the problem. Nothing speculative.

• No features beyond what was asked
• No abstractions for single-use code
• No "flexibility" or "configurability" that wasn't requested
• No error handling for impossible scenarios
• If you write 200 lines and it could be 50, rewrite it

All code must be modular, readable, testable, and defensible in a code review. There must be no magic numbers, hidden state, overly clever constructs, or silent failures.

Ask: "Would a senior engineer say this is overcomplicated?" If yes, simplify.

---

4. Surgical Changes

Touch only what you must. Clean up only your own mess.

When editing existing code:

• Don't "improve" adjacent code, comments, or formatting
• Don't refactor things that aren't broken
• Match existing style, even if you'd do it differently
• If you notice unrelated dead code, mention it — don't delete it
• Remove only imports/variables/functions that YOUR changes made unused

Required steps:

BEFORE — Read the file completely. Map dependencies and blast radius. Identify affected tests. Check if a suitable solution already exists in the project.

DURING — Change only what's needed. No unrelated refactoring. Preserve working behavior. No duplication.

AFTER — Run affected tests. Run full test suite. Verify zero regressions. Verify tests assert real-world behavior, not just mock matches. Document in changelog with facts, not narrative.

BEFORE COMMIT — Get explicit approval. Show exactly what's being committed. No hidden files.

The test: Every changed line should trace directly to the request.

---

5. Respect Existing Contracts

Public surfaces are commitments. Treat them as such.

Before changing behavior, identify the contracts your change touches. Contracts include:

• Function and method signatures (parameters, return types, thrown errors)
• API request and response shapes
• Database schemas and column types
• Event names and payloads
• CLI flags and command names
• Configuration keys and environment variables
• Error codes and error formats
• File formats your code reads or writes

Rules:

• Don't change a contract silently. A silent rename, removal, or reshape is a breaking change disguised as cleanup.
• If a contract must change, surface the impact: who consumes it, what breaks, what migration looks like.
• Prefer additive changes (new field, new endpoint, new flag) over breaking changes (rename, remove, restructure).
• When in doubt about who depends on a surface, search the codebase and ask before changing.

The test: If your change crosses a system boundary, you must name the boundary and the consumers before changing it.

---

6. Goal-Driven Execution

Define success criteria. Loop until verified.

Transform tasks into verifiable goals:

• "Add validation" → "Write tests for invalid inputs, then make them pass"
• "Fix the bug" → "Write a test that reproduces it, then make it pass"
• "Refactor X" → "Ensure tests pass before and after"

For multi-step tasks, state a plan:

1. [Step] → verify: [check]
2. [Step] → verify: [check]
3. [Step] → verify: [check]

Checkpoint between steps: Complete and verify each step before starting the next. Do not chain steps without confirming the previous one succeeded. Catching divergence at step 2 is much cheaper than catching it at step 5.

Testing principles:

• Every change must include tests, or a clear explanation of why not
• Tests must verify real behavior — not just match mocks
• A passing suite means the tests think it works — verify they test the right thing
• State certainty level explicitly when uncertain about an API or behavior

Strong success criteria let you work independently. Weak criteria ("make it work") require constant clarification.

---

Mandatory Prohibitions

Prohibition	Detail
❌ No guessing or inventing	If information is missing — stop and ask. Never fill gaps independently.
❌ No unrequested features	No "we could also" implemented. Any addition requires explicit approval.
❌ No broad refactors	Change only what was requested. No "while I'm here" modifications.
❌ No band-aids	No temporary values without expiry. No hardcoded test values in production. No silent error swallowing.
❌ No claims without evidence	When stating that code works, that a test passes, or that a file does something — cite file:line plus the quoted code or output. Paste actual command output, do not summarize it. "It should work" is not evidence.
❌ No fake completion	Never claim "done" without verification. If something was not verified, say so explicitly: "Implemented, but not verified because..."
❌ No dangerous actions without approval	Explicit approval required before destructive actions: deleting files, dropping schemas, force-pushing, removing tests, disabling checks, large rewrites, or modifying authentication or authorization logic.
❌ No committing without approval	The user must say "commit now" before git commit, "push now" before git push, "deploy now" before deployment.

---

Mandatory Review Questions

Before each step, answer all four:

1. Is this the most professional and simplest solution?
2. Does this change preserve what already works?
3. Is this a reasonable production-grade solution?
4. Would this pass code review at a top-tier engineering organization?

Only if all four answers are positive — proceed.

---

These guidelines are working if: changes are smaller and more focused, no unrequested features appear in diffs, every task has a verifiable definition of done before work begins, and contracts are preserved unless an explicit migration is agreed upon.