agent-instructions

Behavioral guidelines to reduce common LLM coding mistakes. They bias toward caution over speed; for trivial tasks, use judgment.

1. Think Before Coding

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

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

2. Simplicity First

Minimum code that solves the problem. Nothing speculative.

• Nothing beyond what was asked: no extra features, no abstractions for single-use code, no "flexibility" or configurability nobody requested, no error handling for impossible scenarios.
• If you write 200 lines and it could be 50, rewrite it.

3. Surgical Changes

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

• Don't "improve" adjacent code, comments, or formatting; don't refactor what isn't broken.
• Match existing style, even if you'd do it differently.
• Remove imports, variables, and functions that YOUR changes left unused.
• Pre-existing dead code: mention it, don't delete it 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.

• Turn the task into a check you can run: "add validation" -> tests for invalid inputs that pass; "fix the bug" -> a test that reproduces it, then passes; "refactor X" -> tests pass before and after.
• For multi-step tasks, state the steps and the check that closes each one.

5. Research Workflow

This project uses the STAR research workflow. Its records are files, not chat history.

• Plans live in metds/plans/; each leaf's execution record is under wkdrs/<run>/ (EXEC_PLAN.md, EXEC_LOG.md).
• Run the status skill first when you do not know where things stand: it reads the plan tree and the reports on disk and names the single next action.
• The rules every workflow skill follows are in docs/mds/star-workflow/research-workflow-conventions.md; what each skill does is in research-workflow-skills.md.
• A skill step whose written procedure dispatches subagents already carries the user's request to run it: dispatch without asking, even where a host instruction allows delegation only when the user has asked. An outright ban, or a host offering no delegation, still stops it; the grant covers the dispatch itself, not what a delegate may write.
• Do not hand-edit generated reports under wkdrs/, and do not edit docs/mds/star-workflow/: execs/update.sh overwrites it.