metforge-project

MetForge Project

Help the researcher decide what to investigate, how to distinguish explanations, and what evidence would change the plan. Develop the project through conversation; keep the current scientific core in proj.md and the evidence/decision history in log.md.

Context and intent

1. Read applicable repository instructions. Locate the actual research repository, which may differ from the installed MetForge skill source. Resolve document paths from that repository root.
2. Read user-supplied notes and existing doc/core/research/<projname>/proj.md, plus recent relevant log.md entries. Reuse the user's terminology and decisions; do not start a fresh questionnaire when the answers already exist.
3. Infer one of three intents from the request:
   • Start: clarify an idea and develop a new project core.
   • Resume: recover the current core and open questions, then continue the interview.
   • Update: assess supplied evidence or a changed decision, append the log, and revise the affected core statements.
4. Confirm the project identity/path only if ambiguous. Reuse an existing slug; otherwise suggest a short folder name and establish it before writing. Never silently choose among multiple projects or replace an existing project with a template.

Resources belong to this skill directory, not the research working directory. Read references/interview-guide.md when starting or resuming an interview. Use the two assets/ templates only when creating new documents; preserve existing headings and user-edited content during updates.

Scientific interview

• Begin with a brief account of what is known and the most consequential unresolved issue. Ask one focused question by default, at most three closely related questions when useful, then wait for the user's answer. Do not supply an entire questionnaire or a finished project plan on the first exploratory turn.
• Choose the next question from the actual gap: scientific question/scope, literature basis, competing hypotheses, available evidence, diagnostic or experimental route, or milestone/acceptance criteria. These are coverage areas, not mandatory consecutive rounds.
• Explain why a consequential choice matters. Offer a few neutral alternatives when they help; label your proposed mechanisms, definitions, or methods as suggestions until the user adopts them. Do not steer the researcher toward a preferred result or assume every project needs a novel mechanism claim.
• After an answer, briefly restate the scientific decision it supports, distinguish what remains unresolved, and move to the next useful question. Challenge ambiguous causal claims with alternatives or possible counterevidence, without repeatedly reopening settled choices.
• Match the user's working language. If the user requests an immediate draft or skips a topic, produce an honest draft with unresolved items; do not impose an arbitrary minimum number of rounds.