Keep Agent Workflows Simple

Read and apply this skill before planning. Model only the work needed to reach the current outcome. Do not use a board or DAG as a programming language for the agent's thoughts.

If the work changes code, also load keep-code-simple before planning when available.

Write the execution contract first

Define:

• Outcome: one observable result.
• Non-goals: adjacent work excluded from this run.
• Write scope: files, records, services, or decisions that may change.
• Acceptance evidence: commands, artifacts, observations, or explicit decisions.
• Owner: one accountable agent or human.
• Stop condition: finish on evidence; stop on unauthorized scope expansion, unresolved human decisions, unsafe actions, or exhausted limits.

Do not start a persistent goal, create cards, or spawn children merely to discover this contract. Use a bounded read-only inspection first when needed.