Plan.md and Research.md
The two session notebooks Brand Crafter keeps for every onboarding.
For every onboarding session, Brand Crafter keeps two working notebooks in the work area:
Plan.md: the live stage plan and evidence log.Research.md: the open-ended research notebook.
Neither is a finished artifact. Both are working memory the agent reads at the start of every substantive turn.
What Plan.md holds
| Section | Content |
|---|---|
| Active stage | The current stage and its objective. |
| Stage locks | Every lock item with its current state. |
| Approved decisions | Operator-approved facts that have entered the canon draft. |
| Open questions | Unanswered strategic questions, with the stage they belong to. |
| Gaps | Fields the operator could not answer. Each gap names the downstream impact. |
| Risks | Calibration warnings the agent has flagged. |
| Next action | The next concrete move for the agent. |
Plan.md is the ledger. If the session breaks and resumes a week later, Plan.md is what tells the agent where the work was left.
What Research.md holds
| Section | Content |
|---|---|
| Source ledger | Every URL fetched, every document read, with date and confidence. |
| Findings | Public facts the agent has confirmed. |
| Public inferences | Reasonable conclusions drawn from public sources, marked as inference. |
| Contradictions | Where two sources disagree. The agent does not pick a side without operator input. |
| Confidence | How sure the agent is. Low, medium, or high, with the reason. |
| Citations | The provenance trail. |
Research.md is the audit trail. If a future operator asks where a fact came from, Research.md is the answer.
What does not go in either file
The brand kit itself does not live in Plan.md or Research.md. The brand kit lives in the live work area, mounted as a structured artifact with named sections: foundation, strategy, voice, voicePrompt, socialContent, guardrails, summary, brandDesignSystem.
Plan.md and Research.md are scratch. The brand kit is canon.
Where the operator sees them
The two files are not surfaced as a primary operator surface. They show in the work area file tree and in research panels next to the chat. Read them when:
- The onboarding paused and the agent left mid-stage. Open
Plan.mdto see the next action. - A claim in the canon needs a source check. Open
Research.mdto find the citation. - The agent wrote a gap and the operator wants to fix it manually before the next turn.