A free Claude Skill that rewrites founder prose into the register government R&D grant reviewers reward — across NSF SBIR/STTR, DoD SBIR, NIH SBIR, DOE, DARPA, NASA, and similar federal R&D grant programs.
Paste your draft → get back specific, falsifiable, reviewer-ready rewrites — with explanations for every change.
Most founders write R&D grant pitches in marketing register: confident, absolute, feature-focused. Reviewers reward the opposite — describe uncertainty, not capability. This skill teaches the inversion on your own draft.
The skill scans your prose for seven patterns government R&D grant reviewers consistently downgrade:
- Absolute terms (always, every, the best, revolutionary)
- Product-development framing (spin out of, version 2, scale our product)
- "Nobody has done this before" claims
- Vague impact statements
- Over-claiming measurement ("measures critical thinking")
- Missing references on factual claims
- Marketing superlatives (seamless, cutting-edge, next-generation)
For each pattern detected, you get:
- The original phrase quoted verbatim
- One-sentence explanation of why it backfires with reviewers
- A specific, falsifiable rewrite
- A pointer to the underlying framework
The patterns apply across NSF SBIR/STTR, DoD SBIR, NIH SBIR, DOE, DARPA, NASA, NOAA, and similar federal R&D grant programs.
Primarily founders editing their own drafts — short Project Pitches through full Phase I proposals.
Also useful for evaluators: program officers, foundation reviewers, and R&D-savvy angels paste incoming drafts to surface the questions they should ask before approving or investing. The same pattern detection serves both sides — founders read it as "what to fix"; evaluators read it as "what to challenge."
→ Get the skill at elleyuanwang.com/sbir-pitch
Drop your email and you'll get:
- The install command for Claude Code
- A sample passage to try the skill on first
- Launch-week pricing for the full NSF SBIR Pitch Playbook when it ships (the first in the playbook series)
- Occasional notes on what the author is seeing across recent reviewer panels
No spam, unsubscribe one click.
Claude Code (one command)
npx skills add ellewangyuan/grantproof --globalThat's it. The skill installs to ~/.claude/skills/grantproof/ and is available in every Claude Code session. Paste your draft in any conversation and the skill activates automatically.
Useful commands:
npx skills list # show installed skills
npx skills update ellewangyuan/grantproof # update to latestClaude.ai (web)
Claude.ai doesn't yet support skill installation natively. Two manual options:
- Project-based (recommended): create a new Project in Claude.ai → copy the contents of SKILL.md → paste as the Project's custom instructions → start chats in that Project.
- One-off use: paste SKILL.md contents into any fresh conversation, followed by your draft. Claude will follow the skill's behavior for that conversation.
Your draft:
Company A's continuous glucose sensor delivers the most accurate readings of any wearable on the market.
Skill output:
Patterns detected:
- Absolute term: "the most accurate" — unfalsifiable without a defined comparison and reference standard
- Marketing superlative: "on the market" — undefined market segment
- Missing reference: claim of comparative accuracy with no measurement protocol cited
Rewrite:
In a 60-subject pilot under Company A's IRB-approved protocol, our wrist-worn glucose sensor showed a mean absolute relative difference of 9.2% against laboratory venous reference samples in the postprandial window — within the FDA iCGM accuracy threshold (15%) and 28% tighter than the leading FDA-cleared comparator.
Why these changes:
- "The most accurate" → bounded with sample size, protocol, and a specific quantitative metric (MARD). A reviewer can now evaluate whether the claim is testable.
- "On the market" → replaced with a named regulatory threshold and a defined comparator class. Reviewers can locate the work on a map of prior efforts.
- The 28% tighter figure gives reviewers something to either accept or push back on. Counterintuitively, that's better than a confident claim with nothing to verify.
See examples.md for more.
Found a rewrite that helped? Share the screenshot on LinkedIn or send it to your co-founders. The skill's outputs are plain text — they paste cleanly anywhere.
Describe uncertainty, not capability.
This is the single most counter-intuitive rule in grant writing — and the one founders most reliably get wrong.
Founders write R&D grant pitches in marketing register: confident claims, absolute terms, product features. They've been trained for it by every investor pitch they've ever given. Reviewers reward the opposite. Specific, falsifiable, evidence-anchored prose wins. Vague, absolute, unfalsifiable prose loses.
This skill exists to enforce that inversion on your draft, line by line, until it feels natural. The full playbook teaches the underlying mental model — why reviewers reward uncertainty, and how to write prose that survives the panel.
Your draft is processed by Claude (Anthropic). Anthropic's data policies apply. The skill itself doesn't transmit your data anywhere else.
Created by Dr. Elle Yuan Wang:
- PhD, Cognitive Sciences & Learning Analytics — Columbia University
- Served as an NSF SBIR/STTR reviewer (America's Seed Fund)
- Lead Scientist & Advisor, AI-ALOE (National AI Institute for Adult Learning and Online Education)
- Advisor, ASU Online
- Judge / advisory board: $5M XPRIZE IBM Watson AI for Good · $1M XPRIZE-IES Digital Learning Challenge · $4M Learning Engineering Tools Competition
- Coached and supported $21M in collective R&D grant funding over 10 years across federal R&D grant programs
- Strategic roles: Columbia Tech Ventures · MTV Networks · NYC Mayor Bloomberg's Office
A note on what this skill can and cannot do: it enforces a more rigorous writing register that R&D grant reviewers reward. Writing register is one layer of why a proposal gets funded — it does not guarantee approval. Strong writing supports a strong project; it cannot substitute for one.
The skill catches surface-level patterns. The full NSF SBIR Pitch Playbook — the first launch in the playbook series — covers what the skill cannot:
- 12 register patterns reviewers consistently downgrade (with deeper explanation)
- The reviewer's mental model — what the panel actually scores and how
- The 4-section NSF SBIR Pitch storyline order, with notes on how it adapts to other agencies' formats
- 5 worked before/after examples across SBIR-relevant domains
- A 20-point pre-submission checklist
- A 60-minute reviewer mini-training video
Launch-week pricing for waitlist subscribers: https://elleyuanwang.com/sbir-pitch
Issues and PRs welcome. Spotted a pattern that should be flagged? Open an issue. Have a worked example to contribute? Add it to examples.md and PR.
MIT — use it, fork it, share it. See LICENSE.