harness-continue

To resume a long-running codebase session without losing context, use session recovery to summarize the last state, surface untracked files, and verify git identity. This preserves intent and coordination metadata across compactions.

To enforce test-driven development gates during AI agent implementations, run a TDD loop with required task updates, coverage gates, and pre-commit features.json audits. This ensures test discipline before code changes are committed.

The best way to coordinate parallel AI agent teams on a tracked feature is through orchestration hooks that plan, spawn, monitor, and synthesize teammates. This includes dependency management, quality hooks, and mandatory retrospectives.

Context compaction causes lost intent in long-running development sessions because summarized context drops critical metadata. An explicit compaction strategy with persistent context updates and post-compact recovery steps prevents this loss.

Yes, you need a features.json audit before committing changes in a harness-managed project. This pre-commit audit acts as a quality gate to verify task lifecycle updates and maintain metadata-driven handoffs.