review-grant

What it does

The skill runs a six-agent, parallel pre-submission panel review of a grant proposal. It parses input arguments to determine a target funder/program and either uses a specified proposal file or auto-detects the main proposal and accompanying supporting materials. It identifies the main narrative document and relevant appendices, budgets, timelines, biosketches, and other attachments. Then it launches six review agents in parallel, passing all proposal and supporting file paths to each agent, with Agent 6 receiving TARGET_PROGRAM substituted into its prompt. Each agent specializes in a different review dimension (clarity, internal consistency, significance, methods/feasibility, budget/timeline/team/management, and adversarial funding recommendation).

How it works

• Phase 1: Parse Arguments and Discover the Proposal
  • Recognizes target programs/funders: NSF, NIH, ERC, HorizonEurope, major-funder, foundation (case-insensitive).
  • If the first token matches, that becomes TARGET_PROGRAM; remaining text is treated as the main proposal file path. If not, the entire ARGUMENTS is treated as the file path and TARGET_PROGRAM is major-funder. If ARGUMENTS is empty, defaults apply (no file path, TARGET_PROGRAM major-funder).
  • If a file path is provided, use it as the main proposal file; otherwise auto-detect by scanning for likely proposal files with extensions *.md, *.txt, *.tex, *.docx, *.pdf, excluding hidden folders and common system dirs. Prioritize main narrative-like names (proposal, project-description, research-plan, etc.). Identify the main document and extract references to supporting documents (budgets, timelines, biosketches, letters, etc.). Record paths and roles of supporting files found.
  • If binary formats are present, note the limitation explicitly in the final report.
• Phase 2: Launch 6 Review Agents in Parallel
  • In a single message, launch six agents with subagent_type: "general-purpose".
  • Pass the complete list of proposal and supporting file paths to each agent.
  • For Agent 6, substitute the actual TARGET_PROGRAM for every occurrence of TARGET_PROGRAM in its prompt.

Agents cover:

• Agent 1: Clarity, Writing Quality & Compliance Signals
• Agent 2: Internal Consistency, Scope & Deliverables
• Agent 3: Significance, Innovation & Fit to the Call
• Agent 4: Research Design, Methods & Feasibility
• Agent 5: Budget, Timeline, Team & Management Plan
• Agent 6: Adversarial Panel Review & Funding Recommendation

The output is a structured set of agent reports, compiled to inform evaluators on overall fundability and alignment with the specified funder or general standards.