Global Behavioral Guidelines

Applies to all projects. When these conflict with a project's own instructions or existing conventions, the project wins.

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

1. Think Before Coding

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

For non-trivial decisions, work as a thoughtful collaborator: present the options, explain the tradeoffs, recommend a path, and state your assumptions. Prefer practical, simple, portable solutions.

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.
• Proceed when the remaining open questions wouldn't change the approach; otherwise ask — focused questions, only where the answer materially changes the outcome.

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]

Break implementation into the smallest logical, independently verifiable steps that still produce meaningful progress. Complete and verify one step before starting the next. When committing is part of the task, keep the same boundaries: one logical theme per commit rather than one large catch-all commit.

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

5. Branch Before You Build

Never work directly on main/master. Before starting changes in a repo, create a working branch from the current branch:

git checkout -b feature/<short-description>

A long-lived integration branch (e.g. stage) is fine too — the rule is simply that main/master stay clean until a change is ready to merge. Branch names should describe the work, not the date.

A backstop (remote branch protection, a pre-commit hook, or a tool-level guard) should enforce this; this rule is the intent. If a guard blocks you, branch first, then retry.

How hard the tool-level guard pushes is a per-repo decision: off, remind (once per session), ask, or deny. A committed client profile defaults to deny; the built-in and personal defaults are remind. A personal repo that legitimately lives on one branch can override the policy locally with off. The rule above holds regardless of the setting — the policy tunes the interruption, not the convention. A guard that is wrong half the time gets disabled entirely, and then it protects nothing.

When parallel sessions may run against the same repo, worktree before you build. git has one HEAD per working directory, so two sessions (agents or humans) sharing one checkout will land commits on whatever branch is checked out at that instant and tangle histories. Give each feature branch its own git worktree instead of switching branches in a shared checkout:

git worktree add -b feature/<short-description> ../<repo>-worktrees/<slug> <base>

Run one session per worktree; remove it (git worktree remove) once merged. Solo, single-session work can still branch in the main checkout — the worktree is the fix for concurrency, not a tax on every branch. A repo may ship a helper (e.g. scripts/new-worktree.sh) that encodes its base branch and location convention.

Clean up completed work. When you know a feature branch has been merged into main, master, or the repo's integration branch (for example stage):

1. Verify the worktree is clean.
2. Verify the feature branch is fully merged with git merge-base --is-ancestor <branch> <target>.
3. Remove the linked worktree.
4. Delete the merged local branch with git branch -d <branch>.

If the target branch is ambiguous, the worktree is dirty, or ancestry cannot be proven, stop and ask. Never force-delete a branch or delete a remote branch without an explicit request.

6. Operating Safety

Secrets: Never commit secrets or print credential/env values. If you spot a hardcoded secret or key, stop and flag it — don't "fix" it silently.

Irreversible actions: Before rm -rf, dropping DBs/tables, running migrations against non-local environments, force-push, history rewrite, or deleting branches — stop and confirm. The only standing exception is deleting a clean local worktree and local branch after the merged-work cleanup checks above prove the branch is fully merged.

Commits & PRs: Small, logically-scoped commits. Keep the subject line to 72 characters or fewer and describe the commit's overall theme clearly. Add a body only when additional context materially helps the reviewer. Do not add Co-authored-by: trailers or other co-author attribution. Never push or open a PR without an explicit ask.