AGENTS.md

Behavioral guidelines to reduce common LLM coding mistakes. Merge with project-specific instructions as needed.

Tradeoff: These guidelines bias toward caution over speed. For trivial tasks, use judgment.

Shared Rules

Shared repository rules live here:

• ./.shared-agents/common/linting.md

Validation

Before finishing any code task, follow:

• ./.shared-agents/common/linting.md

Do not finish with known safe-to-fix issues in the changed scope.

Commits

Codex adds no default co-author trailer, so it attaches nothing — that is expected and correct. Do not fabricate a Co-Authored-By, session-id, or URL trailer on commit messages.

Model Selection

Keep orchestration, planning, and review on the top model at all times — never downgrade the main agent. Delegate mechanical work to subagents so the orchestrator's context stays clean; the cost win comes from that isolation (and a cheaper worker model), not from starving the worker's reasoning.

Match each subagent's reasoning effort to its task — low for a simple lookup, higher for interdependent changes or debugging — and reach for more effort whenever a cheap miss would be costly to recover from (a failed worker bounces back to the orchestrator at full price). Codex only spawns subagents when asked, so treat this as a standing instruction to delegate.

Route subagents by difficulty — each agent's own model sets its cost, whatever model the main session runs:

• research and hard analysis → researcher (GPT-6 Astra)
• implementing an already-decided plan → executor (GPT-6.1 Sol)
• codebase search → explorer (GPT-6 Luna)

Keystone

Read ~/.claude/skills/keystone/SKILL.md before implementing non-trivial logic, structuring a module, or judging whether existing code reads well — it carries the reading standard all code is written to. React components, pages, and features also take ~/.claude/skills/keystone-react/SKILL.md.

1. Think Before Coding

Don't assume. Don't hide confusion. Surface tradeoffs.

Before implementing:

• State your assumptions explicitly. If uncertain, ask.
• If multiple interpretations exist, present them - don't pick silently.
• If a simpler approach exists, say so. Push back when warranted.
• If something is unclear, stop. Name what's confusing. Ask.

2. 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.

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

3. 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.

When your changes create orphans:

• Remove imports/variables/functions that YOUR changes made unused.
• Don't remove pre-existing dead code unless asked.

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

4. 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 brief plan:

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

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

5. Design Patterns

Reach for an established pattern before inventing structure.

When a problem maps cleanly to a known pattern, use it — don't build a bespoke solution that a reviewer will have to reverse-engineer.

Context7

Context7 is the default source for library and framework documentation.

• Use the context7 tool when you need library or API documentation.
• Use the context7 tool when you need code generation, setup instructions, or configuration steps.
• Do not rely on memory for library versions or APIs when verification is needed.

Shell Tool Preferences

Prefer modern CLI search tools for efficiency:

• Use fd instead of find.
• Use rg instead of grep.

If either tool is missing, inform the user and ask before running an install command.

These guidelines are working if: fewer unnecessary changes in diffs, fewer rewrites due to overcomplication, and clarifying questions come before implementation rather than after mistakes