Grant¶

The grant plugin drafts and reviews NIH and NSF grant proposals with mechanism-specific templates (R01, R21, K99, CAREER, and more), plus a figure-QA skill scoped to grant-specific compliance checks.

The proposal pipeline¶

Writing a proposal is a five-stage managed pipeline, with review findings boomeranging back to an earlier stage rather than being patched in place:

Grant pipeline: NOFO, Aims, Strategy, Self-review, Figure QA, with a boomerang from review findings back to an earlier stage

NOFO: read the funding opportunity as a contract
Aims: one page, three aims
Strategy: significance, innovation, and approach, scored 1-9
Self-review: a study-section simulation via grant-review
Figure QA: resolution, fonts, colorblind-safe palettes via grant-figure-qa

Order is enforced; a self-review that cycles back to an earlier stage is the rigor signal working as intended, not a setback.

Findings from the self-review stage are severity-tagged, and the severity determines where they cycle back to:

Grant review boomerang: critical findings cycle back to Aims, major findings to Strategy, minor findings get an in-place edit with no cycle

Critical: revise the Aims
Major: tighten the Approach
Minor: edit in place, no cycle needed

How NIH scores a proposal¶

grant-review's NIH rubric mirrors how an actual study section scores a proposal: five criteria, each 1-9, rolled into one integrated impact score.

NIH study section scoring: Significance, Investigator, Innovation, Approach, Environment, each scored 1 (exceptional) to 9 (poor), rolled into an integrated impact score

A 1 in any single criterion does not save a 5 in Approach; reviewers tie everything back to the integrated impact score, not the best individual criterion. (As of the 2025 Simplified Review Framework, RPG mechanisms score three factors, Importance, Rigor & Feasibility, and Expertise & Resources, rather than the five shown here; F fellowships and T training grants follow their own separately revised frameworks. See the grant-review skill for the current mechanism-specific rubric.)

Skills¶

grant-writing: research strategy guidelines, writing style, budget justification, and resubmission response, with mechanism-specific templates
grant-review: the NIH/NSF study-section simulation described above; a thin-dispatch skill with a Claude-bundled fresh-context agent, a Codex agent template, and a Copilot plugin-agent template; when no agent is configured, the skill runs the same reference procedure inline
grant-figure-qa: checks figures for resolution, accessibility, and NIH/NSF compliance, following the same dispatch pattern as grant-review

Try it¶

"Write the significance section for an R01 on motor cortex"
"Review my R21 proposal at proposal.pdf as an NIH study section"

Learn more¶

The Agentic Research Course week 6, "Grant Proposal Writing," walks through NIH/NSF proposals hands-on.