history-hygiene

To prevent stale decision logging in agent workflows, record only confirmed final outcomes in history.md instead of intermediate requests. This ensures future agents read accurate decision context as ground truth without being poisoned by drafts or rollback-later notes.

Outcome-only history logging for collaborative documentation records what actually happened rather than what was initially requested. It applies a confirmation-before-write principle to history.md, ensuring entries reflect the latest final decision instead of intermediate states.

To update history.md when a decision is reversed in an agent workflow, write a single confirmed final outcome entry reflecting the latest decision immediately. This immediate correction ensures the one read equals one truth principle holds for future agents.

Avoiding intermediate-state writes to history.md during agent workflows prevents poisoning decision context with drafts or for now states. Future agents read history as ground truth, so logging unconfirmed requests directly causes bad decisions when earlier requests are later reversed.

The history-hygiene skill works without external dependencies for decision logging, requiring no components or packages. It directly maintains history integrity by applying confirmation-before-write rules to your history.md file within agent-driven workflows.