grant-compasslisted
Install: claude install-skill rx290/grant-compass
Token budget matters here. Do all deterministic fetching/parsing/scoring in the
`grantcompass` CLI (local Python, no LLM tokens). Only read the final compact
`output/report.md` back into context, never raw HTML, raw JSON, or full page
dumps. Do not pad the final answer with encouragement or filler; report the
table and one line on grade fit and deadlines, nothing else.
Steps:
1. Run `python -m grantcompass init` (from the repo root). If it just created
`config.local.yaml` (or if `applicant.name` is still the placeholder
`"Your Name"`), **ask the user directly, in the conversation**, before doing
anything else:
- Which country/countries do they want to target? If they don't say,
default to all of Europe and tell them that's what you're doing.
- Which degree level(s): Bachelor's, Master's/MS, and/or PhD?
- Their field/keywords, and their grade (GPA/CGPA/percentage, whatever
scale they use) and its scale.
Write their answers into `config.local.yaml` yourself (it's gitignored,
never guess or invent values, and never commit it). Then continue.
2. Run `python -m grantcompass professors`, a deterministic OpenReview/OpenAlex
pass, no LLM involvement. Skip this entirely if `degree_target` is
Bachelor's only, undergrad applicants don't need PI matching.
3. Program/funding listings have no stable public API, so this step needs your
own web access:
- Read `src/grantcompass/sources.yaml`.
- Filter entries to those matching the applicant's