agent-harness

Decomposes a project milestone into bounded, verifiable feature-list entries suitable for the planner/coder/reviewer cycle in the agent harness.

Handles bounded bug-fix sessions in the long-running harness by reproducing, isolating, fixing, verifying, and recording defects with explicit blocker and issue tracking.

Applies a structured evidence-based review checklist for bounded session results, validating scope, correctness, verification, artifact consistency, and resumable repository state.

Implements a selected bounded feature task within the long-running harness, respecting current-task scope, verification requirements, artifact updates, and clean-state handoff rules.

Evaluates whether a bounded implementation or bug-fix session introduced regressions, using task-aware verification evidence, repository state checks, and explicit risk recording.

Establishes a reproducible repository bootstrap path for long-running agent sessions, including execution entry points, state artifacts, smoke readiness, and baseline environment normalization.

Breaks large or ambiguous work into bounded, verifiable session-sized tasks that can be selected, implemented, reviewed, and handed off safely within the harness.

Enforces the repository's verification gate for bounded sessions by evaluating required test scope, execution evidence, pass/fail status, and gate impact on task approval.