Model Audit Tie-Out

Skill Configuration

User Context Preflight

Before searching connectors, retrieving evidence, or drafting output, run python3 skills/user-context/scripts/user_context_preflight.py with the shell working directory set to this plugin's root, and follow the returned saved_context, source_category_plan, and next_action. Set the working directory before the first attempt; do not probe alternate relative paths. Missing context must not block the requested workflow. Do not initialize state or run onboarding during ordinary workflow work.

If next_action.id = "offer_orientation" and the parent router has not already handled it, complete the requested work first and append its one-line optional setup offer once.

Internal Support

When the audit requires evidence control, generic data cleaning, dashboard rendering, style application, or sector context, route that support through the visible public-equity-investing router and its bundled internal playbooks. This visible skill remains the owner of the model-review deliverable.

Deliverable Intake

Apply the presentation-surface precedence in ../../shared/deliverable-intake-policy.md. This workflow's natural artifact is a polished standalone HTML model-audit report, while direct remediation preserves the source workbook format. Do not choose chat-only output unless the user explicitly requests a lightweight response.

Before source gathering or analysis for a new standalone reader-facing hero deliverable, load ../../shared/deliverable-intake-policy.md and use its adaptive request_user_input preflight for materially unresolved format, depth, audience/use, or focus choices. For a substantive standalone review of an existing model or explicit HTML model-audit request, the default resolves the presentation surface to a polished standalone HTML model-audit report unless the user requests an alternate surface, a quick/no-file answer, remediation output, or a standardized dashboard. In interactive runs, ask only remaining material choices such as depth, audience/use, materiality, or focus; in non-interactive runs, default to the HTML model-audit report and Full working analysis while disclosing those assumptions outside the artifact. Reuse resolved preferences in downstream steps; when acting only as input to an owning workflow, do not re-prompt.