Runelace
A pre-deployment workbench that maps source runbooks to typed action candidates, exposes unsupported assumptions and routes every contract through owner review and simulation.
Enterprise support teams may have human runbooks that mix policy, judgment, advice and deterministic actions. The supplied research found open-source runtime and contract-specification projects, but no commercial product combining prose ingestion with linting against an already deployed support agent's declared capabilities. Three interface references were not verified in the earlier stage.
Runelace preserves each source passage and proposes whether it is informational, advisory, decision-dependent or executable. For executable candidates it drafts inputs, authority, preconditions, tool action, success evidence, failure handling and escalation. A runbook owner corrects the classification and approves the contract; deterministic lint and simulation then compare it with a versioned capability manifest.
A model classification is not policy truth. A declared tool does not prove permission, correct configuration or safe runtime behavior. A passed lint or signed report is not production readiness, security assurance or deployment authorization. Source text, interpretation candidate, approved contract, capability claim, simulation result, release approval, execution, destination readback and customer outcome remain distinct.
The first release should cover one read-only support runbook and a manually verified capability manifest. It must not execute actions, ingest secrets, infer refund or account authority, silently convert human judgment into rules or expose internal support knowledge beyond its approved tenant.
Support automation, knowledge-management or AI-governance leader preparing existing runbooks for controlled agent deployment
The supplied signal describes enterprise rollback and growing demand for governance and audit.
Parsing and deterministic lint scale predominantly through software after manifest maintenance.
The current shift toward agent governance explains timing but not a strong historical barrier.
The supplied research confirms adjacent runbook runtimes and contract standards while leaving a reviewed-set gap for source-linked translation plus capability linting.
The buyer profile is incomplete, three interfaces were unverified, prose classification requires substantial review and the segment gap does not establish a structural incumbent barrier.
Discussion
No comments yet — be the first to weigh in.