AI-Assisted Workflow Extensions

These draft proposals describe optional AI-supported requirements inspection. They are based on the proposed concept and public S-CORE process definitions. They do not establish standards compliance, tool qualification, or coverage of a particular safety level.

The Bazel documentation bundle publishes these draft nodes alongside the existing process. Role assignments are proposed responsibilities, not recorded approvals. The executable demonstration is described in Execute the Requirements Inspection Demonstrator.

The workflow augments wf__monitor_verify_requirements without replacing the existing manual inspection. Proposed tool evaluation and monitoring are described in AI Tool Management.

AI-Assisted Safety Requirements Inspection		status: draft
tags: requirements_engineering version: 1 contains: gd_chklst__req_inspection needs input: wp__requirements_comp, wp__requirements_feat outputs: wp__requirements_inspect responsible: rl__committer approved by: rl__committer supported by: rl__safety_manager
An engineer selects feature or component requirements, context, and the applicable public S-CORE inspection checklist. Before sharing input with an AI service, the engineer verifies that sharing is permitted. Roles match Monitor/Verify Requirements, which this workflow augments: the committer remains responsible and approving, with the safety manager supporting as moderator, consistent with Work product Inspections Co.... This also applies to re-inspection triggered by change or release review; no separate reusable sub-workflow is defined for that case. The AI fills the checklist content directly in the exact column structure of Requirements Inspection Che... (Review ID, Acceptance Criteria, Guidance, Passed, Remarks, Issue link), parsed from that template rather than paraphrased, without changing requirements or inventing checklist IDs. Each entry has a yes / no / n/a / not_assessed verdict (missing context is never silently treated as n/a) and full remarks; an issue link is mandatory for every no. The result is presented as one editable table; the original judgment remains preserved separately. Scope: The existing wf__monitor_verify_requirements includes wp__requirements_stkh, wp__requirements_feat, and wp__requirements_comp. This AI variant assesses feature and component requirements only. Stakeholder requirements may be relevant parent context, but their distinct checklist profile is not implemented and their text is not retrieved automatically unless the engineer explicitly supplies it as context, in which case linkage/completeness items may be assessed instead of marked not_assessed. Proposed review guardrail: After presenting a validated draft, the agent ends its turn. The reviewer independently checks the assessment, evidence gaps, parent traceability, and consistency, and may edit the checklist or request corrections; the agent validates and presents every changed revision, then stops again. There is no local chat-confirmation command: the engineer copies the checked checklist into the project’s real inspection document (for example doc__<feature>_req_inspection), commits it, and the normal Git/GitHub review of that change is the only approval this workflow recognizes. Editing, Ready, positive feedback, and tool permissions are not that approval. Proposed constraints: Validation rejects entries outside the feature/component checklist profile, unknown checklist IDs, or checklist text diverging from Requirements Inspection Che.... The reviewer checks the AI assessment independently against the requirements and applicable checklist as part of the normal PR review; there is no separate local confirmation step to substitute for that review. Drafts are distinguished from accepted evidence: nothing is written to the project’s inspection document until the engineer does so themselves. Human approval alone does not establish tool qualification. Tool evaluation under the existing tool-management workflows (see AI Tool Management) must be assessed for the intended use before adoption. Model, prompt, configuration, or usage changes trigger impact evaluation.
wf__ai_req_safety_inspect		Workflow