uinaf Slopprep
Make the declared repository and runner dependable for the requested task. Default improvement target: B; C is a checkpoint. Inspection reports state and gaps without expanding into repairs.
Establish scope and evidence
Identify the requested task classes (implementation, qa, or both), runner,
and target grade. Trace the lifecycle through tracked task owners, including
hidden CI and delegated packages: a package verify may cover only tooling
while another task graph owns product proof.
Check prerequisites, cost, credentials, and state changes before unfamiliar commands. Run authorized checks without asking again; inspection alone does not authorize bootstrap, live or paid checks, repairs, or teardown. Read resource ownership before starting or stopping resources.
Grade repository, runner, and exercised evidence separately with grading.md. Unavailable or unsafe checks are gaps, not failed executions.
Choose the relevant guidance
- Agent guide structure, routing, authority, cross-model behavior: agent-guidance.md.
- Verification coverage, real surfaces, failure evidence: verification-contract.md.
- Slow gates, affected selection, caching: fast-portable-execution.md. Measure unchanged, relevant-change, warm-full, and cold-full paths before optimizing.
- Boot, doctor, resource ownership, identity, isolation, recovery: setup-patterns.md.
- Unattended workflow proof, repeated trials, A-grade reliability claims: autonomy-evidence.md.
Improve and finish
Extend the existing manifest, compiler, test framework, or typed CLI; a plain repository script may suffice. Keep private human context and harness settings with their owners; never copy or print secrets.
Continue authorized repairs through proof of the changed lifecycle on success and a safe failure. Preserve primary status, classified diagnostics, recovery instructions, and task/attempt artifacts. Use the owner's affected checks and reuse passing proof until a change, failure, or concrete concern invalidates it.
Report repository/runner grades before and after, evidence level, strongest outcome, first missing automation transition, and remaining gaps with owners. Finish at the requested outcome or an evidenced blocker; readiness work alone does not authorize shipping or submitting changes.