| name | niw-evidence-finder |
| description | Searches authoritative U.S. government sources for evidence that a specific EB-2 NIW proposed endeavor advances a named national priority, and rates honestly how strong the result is. Use when the user asks for evidence of national importance, for government sources or programs supporting their endeavor, for help with Prong 1 evidence, or when drafting an endeavor statement or petition section that needs sources. Sorts what it finds into endeavor-specific, sub-field and field-level tiers, and says plainly when nothing better than field-level framing exists rather than dressing it up. Requires a working search tool, so it needs an environment with network access. |
| license | Apache-2.0 |
| allowed-tools | WebSearch, WebFetch, Read, Write, Grep, Glob |
| compatibility | Requires network access and a search tool. Works in Claude Code. On claude.ai it depends on whether network access is enabled for your account. It cannot work through the Claude API, where skills have no network access. |
niw-evidence-finder
Find evidence that this specific endeavor advances a named U.S. national priority, and be honest about how strong what you found actually is.
Read references/national-importance-sources.md first.
Before you start: can you search?
This skill depends on a working search tool. If you have none, stop and say so. Do not answer from memory: program names, report titles and agency initiatives change, and a plausible-sounding one that does not exist is worse than nothing. Tell the user this skill needs an environment with network access, and offer to continue without sourced evidence by naming the kinds of source that would help.
The problem this solves
Prong 1 asks about the endeavor, not the field. Evidence that a technology is important, that a market is large, or that a public problem is serious is framing. Records resting on it are refused for exactly that reason.
So the question is narrower: is there a named national priority that this endeavor's specific outputs would advance, and is that advance plausible from what the endeavor actually is?
Workflow
Evidence search:
- [ ] 1. Read the endeavor and extract what to search for
- [ ] 2. Run multiple targeted searches
- [ ] 3. Tier each source and record it in the memo file as you go
- [ ] 4. Rate sufficiency honestly
- [ ] 5. Finish the memo
Step 1: what to search for
Read the endeavor from niw-case/endeavor.md or from the user. Extract the specific problem, technology, method, population and outcome. Search for those, not for the field.
If what you are given is a field rather than an endeavor ("advancing AI for healthcare"), say so first. No search will fix a field-level endeavor, and finding Tier C material for it will make the petition worse by encouraging reliance on it.
Step 2: search
Run several targeted searches across the specific problem, the technology, likely program names and the agencies that would own it. Prefer official U.S. government sources: the White House, executive agencies, Congress, GAO, national laboratories, agency strategic plans and funding announcements. News articles and academic papers are not primary evidence of a national priority.
Step 3: tier and record
For each source: full official title, issuing body, tier with one phrase on why, the concrete connection to the endeavor's specific outputs, how it should be used, and a short quotation copied exactly from what you actually retrieved.
- Tier A, endeavor-specific: a program, funding opportunity, contract, report, strategic plan or regulation targeting the specific problem, technology, method or outcome.
- Tier B, sub-field priority: a national strategy naming the endeavor's specific sub-area.
- Tier C, field-level framing: fact sheets, mission statements, technology lists, market reports. Context only. Label it as such.
If the honest connection is only field-level, say so, even for a source you would rather present as Tier A.
Step 4: rate it
Strong is two or more Tier A. Moderate is one Tier A or two or more Tier B. Otherwise Weak.
If it is not Strong, say what to do instead, and do not suggest finding more Tier C. The useful next step is usually letters from people at the specific programs named, or documenting adoption directly.
If nothing above Tier C genuinely fits, say that plainly. Do not invent a Tier A source and do not promote a Tier C one to improve the rating.
Step 5: write the memo
Create niw-case/research/national-importance.md using the Write tool, ordered Tier A first, with the retrieval date. Later skills draft only from sources in this file and cannot see your reply, so a source described only in the conversation will never be cited.
Searching can consume a long session. Write the file as soon as you have your first recorded source and update it as you go, rather than composing it at the end. A run that stops early then still leaves the petitioner with what you found, instead of nothing.
Write the file first, then confirm it is there, and only then describe it. Do not tell the petitioner it was saved unless you saved it.
Rules
- Search. Never answer from memory.
- Never write a URL you did not retrieve. Record the one the search actually returned, or record none.
- Never place text in quotation marks unless you copied it from the retrieved source. Where no verbatim quotation is available, write that instead of composing one.
- Never invent report numbers, docket numbers, order numbers or program names.
- Do not upgrade a tier to make a case look better. The rating is the useful output.
- No em-dashes. Short plain sentences.
Reference files
What this skill does not do
- It does not assess the case or the endeavor's merits.
- It does not draft anything.
niw-endeavor-statement and niw-petition-letter draft from what this produces.
- It does not search open case law or cite legal authority. Its subject is government priorities, not law.
