session-harness-creator

Automated verification for agentic coding tasks is achieved by generating a task-specific verification harness containing disposable runtime environments and verifier scripts. This structured setup defines clear pass/fail criteria to ensure implementation success before and during development.

A task-specific verification harness is a disposable, task-local directory containing a stable runtime and verifier scripts. You need one for high-uncertainty tasks with significant blast radius, or when requiring automated runtime verification for complex features like data migrations.

Bounded-loop convergence for implementation logic works by having the agent iteratively execute a task, receive automated verifier feedback, and update its progress and scratchpad. This loop continues until the agent converges on a correct solution.

The verification harness requires bash, jq, and git dependencies to function. These tools support the creation of the task-specific directory, manage the stable runtime, and execute the verifier scripts needed for iterative testing and state maintenance.

The best way to ensure agent success on security-sensitive logic is to generate a disposable harness with automated runtime verification. This provides failure mode analysis and bounded-loop execution, allowing the agent to iterate against defined pass/fail criteria until convergence.

You should avoid generating a disposable verification rig for low-uncertainty tasks or scenarios with minimal blast radius. If an implementation task does not require automated runtime verification or iterative convergence, the overhead of maintaining state files and verifier scripts is unnecessary.