med-autogrant

name	med-autogrant
description	Use when Codex needs Med Auto Grant (MAG) to plan, author, critique, revise, or package a medical grant application, including funding-call intake, fundability strategy, specific aims, proposal sections, rebuttal, and a submission-ready local package. Do not use for research-paper production, generic document formatting, patient care, or irreversible submission to a sponsor portal.

Med Auto Grant

Canonical OPL Agent and Package id is mag; med-autogrant is the repository, distribution, plugin, and Skill locator. Use only the installed OPL-generated MAG actions as user entrypoints.

Admission

• Admit MAG when the requested outcome is a grant application or grant-specific strategy/review/package tied to a funding call or an identifiable grant workspace.
• Route research study and paper production to MAS, visual presentations to RCA, and generic document formatting to the relevant document capability unless they are subordinate artifacts of the grant workflow.
• Bind the current funding call, applicant/project context, exact workspace/input refs, task intent, and any accepted upstream decisions. Missing call material is a typed source gap for intake, not permission to invent sponsor requirements.
• Keep sponsor-portal submission, certification, signatures, and other irreversible external actions behind explicit human authority. submission-ready means a local package passed MAG gates; it does not mean submitted.

Action Routing

Use one of the three public actions declared in contracts/action_catalog.json:

• open_grant_user_loop: default end-to-end entry. Use for a new or continuing funding-call workflow spanning intake, fundability strategy, aims/structure, proposal authoring, and review/rebuttal.
• build_direct_entry: enter proposal_authoring directly only when the current workspace already contains sufficient accepted call, strategy, and structure refs for the requested bounded authoring task.
• build_submission_ready_package: enter package_and_submit_ready only when the user explicitly requests a local submission package, provides an output directory, and the human submission_ready_export_gate can be satisfied.

Do not expose repo-local CLI commands, MedAutoGrantDomainEntry, domain-handler dispatch/export, private scanners, or source-purity wrappers as user entrypoints. They are implementation details or authority targets behind OPL-generated surfaces.

Default Workflow

1. Select open_grant_user_loop unless accepted upstream refs justify a bounded direct authoring or package action.
2. Keep one funding-call identity and OPL StageRun lineage across call_and_candidate_intake -> fundability_strategy -> specific_aims_and_structure -> proposal_authoring -> review_and_rebuttal -> package_and_submit_ready.
3. At each Stage, read the current grant artifact, route/quality findings, sponsor requirements, owner decisions, and exact source refs before writing.
4. Let the decisive Codex Attempt route forward, repeat, or return to the highest owning Stage; OPL materializes the transition and MAG retains grant truth and verdict authority.
5. Build the local submission package only after current review evidence, package requirements, output target, and explicit human export approval are present.

Quality And Hard Stops

• Treat any readable grant draft, fundability judgment, review finding, negative result, partial package, failed attempt, or diagnostic as progress and valid route context.
• Retry, review, and repair counts are quality budgets. On exhaustion, preserve the best artifact and continue as completed_with_quality_debt; do not loop until validators clear.
• Static transition tables, schemas, normalizers, validators, generated receipts, provider completion, and tests cannot overrule the decisive semantic route or establish grant-ready / submission-ready.
• Quality debt closes fundable, grant-ready, submission-ready, export-ready, accepted, and production-ready claims while allowing later declared stages to consume the artifact.
• Only executor unavailability, wrong-target identity/currentness, permission/safety/authority boundaries, missing protected credentials/material, irreversible external submission, or an explicit owner/human decision may hard-stop progress.

Output Expectations

• Return the selected public action, funding-call/workspace/input identity, exact grant artifact refs, review and package findings, owner/human-gate refs, typed blockers, remaining quality debt, and next owning Stage.
• Distinguish strategy, aims/structure, proposal draft, reviewed candidate, and local submission-ready package. Never describe a package as submitted without exact external human-authorized evidence.
• Preserve MAG-owned fundability, quality, artifact/package, memory, and owner-receipt authority. OPL-generated status or scanner conformance proves transport/structure only.
• Use OPL-generated status/workbench surfaces for progress and resume; do not create a second MAG runtime, queue, session, or lifecycle model.

References

• contracts/action_catalog.json
• agent/stages/manifest.json
• contracts/stage_quality_cycle_policy.json
• contracts/owner_receipt_contract.json
• contracts/runtime-program/current-program.json
• docs/project.md
• docs/status.md