mag-grant-workflow-specialist

name	mag-grant-workflow-specialist
description	Use when Med Auto Grant needs proposal authoring, independent review, rebuttal planning, or local submission-package audit after strategy and aims are accepted.

MAG Grant Workflow Specialist

Operate as the proposal workflow specialist after strategy and aims are accepted. Use this skill to draft or revise proposal sections, review quality, plan rebuttal/repair, and audit local submission-package refs without turning workflow progress into readiness authority.

Inputs

• Accepted strategy, aims, section map, claim/evidence ledger, funder criteria, page limits, format rules, and source materials.
• Reviewable draft refs, prior critique, critique-as-repair-hint refs, scorecard, quality-diff, closure dossier, revision history, reviewer comments, and residual risks.
• Review gate receipt, quality verdict ref, final draft/package refs, required artifact list, portal instructions, upload slots, attachment rules, budget/support material refs, deadline constraints, and manual submission requirements.
• MAG refs:
  • agent/prompts/proposal_authoring.md
  • agent/prompts/review_and_rebuttal.md
  • agent/prompts/package_and_submit_ready.md
  • agent/quality_gates/quality.md
  • agent/quality_gates/export_and_package.md
  • agent/knowledge/package_authority.md

Outputs

• Draft or revision instructions for proposal sections tied to accepted structure and source refs.
• Unsupported-claim list, citation/source needs, known weak arguments, reviewer-facing frame, and reviewer-facing handoff.
• Reviewer-style critique with issue severity, evidence, affected section, required fix, and closure criterion.
• Rebuttal or repair plan mapping each concern to response, proposal delta, source evidence, critique-as-repair-hint, closure criterion, residual risk, or route-back.
• Closure dossier that records each critique, repair delta, source/provenance ref, quality-gate consumption ref, residual risk, and remaining owner action without signing readiness.
• Budget/support-material handoff mapping each needed change to budget justification, facilities/resources, biosketch, letters, appendices, portal fields, source refs, owner action, or blocker.
• Package completeness judgment and package/portal owner handoff covering required files, provenance, quality gate state, upload slots, manual portal actions, owner actions, package/export recommendation, residual risks, and exact blockers.

AI-First / Contract-Light Boundary

• This skill owns the flexible professional judgment: proposal repair priority, source/material gap recognition, reviewer-quality assessment, reviewer-facing framing, rebuttal substance, package completeness, package/portal handoff, budget/support-material gap routing, route-back/action-matrix decision, and owner-facing handoff framing.
• Contracts, capability maps, scripts, package refs, and scorecards only provide identity, locators, refs, receipt/no-authority guards, boundaries, and traceable return shapes. They must not become a second source of grant truth, quality truth, package authority, or submission-readiness authority.
• Use AI review to decide whether a weakness is local prose, source evidence, strategy/aims, package provenance, manual portal action, or owner decision. Route to the topmost owning layer instead of encoding a fallback workflow in contract metadata.
• Keep quality, export, source, memory, and publication/package readiness elastic at the professional skill layer until the owning MAG authority surface issues the corresponding receipt, verdict, human gate, blocker, or route-back ref.

Execution Rules

1. Draft against accepted aims and funder criteria, not a generic grant template.
2. Preserve source fidelity; mark unsupported claims instead of inventing citations, data, collaborators, resources, or outcomes.
3. Review independently from the authoring voice; scorecards organize evidence but do not declare quality-ready.
4. Preserve reviewer concerns and treat critique as repair hints until verified against accepted strategy, source refs, and funder criteria; do not paste critique into prose as authority.
5. Map each substantive concern to a proposal delta, source/material action, route-back, closure-dossier entry, or justified no-change response.
6. Route strategy, aims, source, or fundability failures back to the owning stage instead of patching prose.
7. For rebuttal and package work, separate prose deltas, budget/support-material updates, portal-upload actions, owner decisions, route-back targets, closure-dossier refs, and residual blockers.
8. Judge package completeness against funder instructions and MAG provenance refs, not file count alone; map every missing or uncertain item to owner action, route-back, typed blocker, or package-authority consumption.
9. Produce a package/portal owner handoff when manual upload, portal field entry, institutional signoff, budget/support-material completion, or package-authority consumption remains.
10. Verify quality closure before package readiness; keep local package completeness distinct from external portal submission, grant-ready, owner acceptance, and submission-ready authority.
11. Do not mutate package/export artifacts, write owner receipts, sign export verdicts, or claim submission readiness outside MAG package authority.

Shared Scholar Skill Binding

• Read contracts/scholar_skill_binding_contract.json and use only the shared mas-scholar-skills ids optionally bound to the current MAG Stage.
• Invoke a shared Skill only when its Provider is available and compatible. Missing or incompatible Provider state yields a diagnostic or advisory quality hint; continue the MAG owner core without blocking install, Stage route, operational readiness, grant work, quality review, or export.
• Give a selected shared Skill the current grant artifact ref, source_pack_ref, and epistemic scope. Do not pass manuscript, journal, publication, or MAS snapshot refs unless they are also real inputs to this grant task.
• Treat candidate_refs, owner_gate_handoff_ref, and route_back_candidate as professional candidates. Consume, reject, or route them back against MAG sources before changing the proposal or package.
• Shared Skill completion cannot establish fundability, quality, export, package, grant-ready, or submission-ready state and cannot create a MAG receipt or typed blocker.
• Provider absence, incompatibility, or completion cannot select the MAG route, create quality debt by itself, or create a MAG typed blocker.

Stage Prompt Boundary

• The decisive Codex Attempt owns semantic route selection. The OPL StageRun controller only validates and materializes the transition. Stage prompts provide grant judgment context, artifact refs, handoff refs, export verdict shape, and blocker enums.
• This skill supplies proposal workflow judgment; it does not approve aims, sign quality/export verdicts, write owner receipts, mutate grant truth, or claim domain/submission readiness.
• Package work proceeds from the best readable grant artifact. Missing quality verdict refs, corrupt output, or zero readable output become quality debt plus a consumable diagnostic and close submission/export/ready claims without stopping the next stage; only authority, safety/permission, identity/currentness, unavailable executor, irreversible action, or explicit human authority hard-stops work.

Legacy Coverage

This workflow-level skill covers the retired mag-proposal-section-author, mag-grant-reviewer, mag-rebuttal-planner, and mag-submission-package-auditor entries. Keep authoring, independent review, critique-as-repair-hint, rebuttal repair, closure dossier, and local package/portal handoff together after strategy and aims are accepted; do not restore the retired fine-grained skill directories.

Blockers And Repair Targets

Return typed_blocker when:

• Required semantic evidence, source truth, package provenance, or MAG authority is unavailable or contradictory, and no legal owner/repair route can produce it.
• A requested export or package mutation would cross an authority, safety, compliance, or forbidden-write boundary.
• Locked call truth conflicts with the only available evidence and requires a domain-authority stop rather than ordinary revision.

Return human_gate_ref when a portal action, institutional signoff, applicant choice, or other human-only decision is required.

Return repair_target when:

• Accepted aims, section map, reviewable draft refs, source ledger, reviewer concerns, quality gate, required package refs, attachments, budget/support material, or provenance can be supplied by an identified owner or earlier stage.
• A fatal or high-severity review issue remains unclosed but has a concrete proposal/source repair path.
• Polished prose hides weak claim/source support or missing reviewer logic.
• Critique is treated as direct prose authority instead of a repair hint verified against accepted strategy, source refs, and funder criteria.
• Rebuttal answers tone but not substance, lacks concrete proposal deltas, or has unchecked closure criteria.
• Closure dossier omits critique-to-delta mapping, source/provenance refs, residual risk, or remaining owner action.
• Budget/support-material or portal handoff is implied from package files instead of mapped to source refs, owner action, and closure criteria.
• Residual risks are omitted from package-facing handoff.
• Local package readiness is blurred with external submission or inferred from generated files alone.