aigon-research-create

aigon-research-create

Run this command followed by the research topic name.

aigon research-create $ARGUMENTS

This creates a new research topic in ./docs/specs/research-topics/01-inbox/.

Tip: If you are running this command in your own Claude Code session (no --agent flag), press Shift+Tab now to enter plan mode before drafting — this keeps your session read-only while you frame the research brief.

Important: Create the document only — do NOT conduct research

Your job is ONLY to create a research topic document. You are NOT researching, investigating, or answering questions. You are writing down what SHOULD be researched later by a different agent (or yourself in a later step).

Do NOT:

• Read source code to investigate the topic
• Search the web for answers
• Write findings, recommendations, or conclusions
• Fill in answers to the questions you define
• Read or explore the codebase even to phrase questions

Do:

• Ask the user what they want to learn (if unclear from the topic name)
• Write a clear Context section explaining why this research matters
• Write focused Questions to Answer as a checklist (unchecked - [ ])
• Define Scope (in scope / out of scope) to keep the research bounded
• Optionally note any Inspiration or starting-point references the user mentions

The output should be a short, well-structured document that frames the research — not the research itself. Think of it as a brief for another agent.

Deepen

Framing only: do not read code, open or investigate references, search the web, or attempt to answer any research question. This rule overrides any instinct to investigate. Your job in this step is only to sharpen the brief for a later research agent.

1. Apply the gate. Inspect the raw invocation arguments. If they contain --quick, skip Deepen and continue to writing the brief. Otherwise run aigon config get deepen.enabled; skip Deepen only when its effective value is false. Remember whether the command reported the built-in default so the final response can include the default-only hint below.
2. Use only supplied framing. Read the bare-bones brief, topic name, planning conversation, and the names or descriptions of user-provided references. Do not open those references. Build an internal coverage map in this order: Context (especially why now and what decision the work should inform) → Questions to Answer → Scope (in and out) → optional Inspiration → complexity:. Mark anything already resolved and do not ask it again.
3. Ask the highest-leverage unresolved framing question. Ask exactly one question per message and include a separate line in this form: Recommended framing: .... When the question offers discrete framing alternatives, enumerate them as short lettered options — a) …, b) … — and name the chosen letter in the recommendation (Recommended framing: a) ...), so the user can answer with just a letter. The recommendation must be a framing choice the user can ratify or override, never a proposed research answer. Normally ask 2–5 questions. Never exceed 6 unless the user explicitly asks to continue, and stop earlier once every material section is draftable. Do not ask about Inspiration merely to fill the template.
4. Maintain the brief as answers arrive. Record ratifications and overrides. If the user says "I don't know" about a substantive answer, turn the unknown into a clear, non-duplicative unchecked item under Questions to Answer. If they do not know a framing preference, use the recommended framing and state that assumption visibly in the relevant Context or Scope prose; never invent a research answer. If the user says enough, stop, or an equivalent exit phrase, stop asking immediately and write the best brief supported by what you have.
5. Keep questions researchable and useful. Each question should be answerable, evidence-seeking, and relevant to a decision. For comparative or empirical questions, name the case, alternatives, and evidence or metric where applicable. For exploratory or qualitative questions, seek mechanisms, constraints, user needs, or failure modes; do not invent a metric merely to make a question look empirical.
6. Finish the framing. Write the brief, then infer complexity: last using the rubric below. In the final response, explain the chosen complexity in one sentence, but do not add a rationale section or rationale sentence to the research brief. If Deepen ran and aigon config get deepen.enabled reported the built-in default, also include exactly this concise hint: Skip next time with --quick; disable everywhere with aigon config set --global deepen.enabled false. Omit the hint when project or global configuration supplied the value.