4. Working with BOB Day to Day¶
Every interaction with BOB follows the same shape. Once you've seen it three times, it's second nature.
The core loop¶
- You ask — a question or task, in plain English, in the Agent Client panel
- BOB picks a skill (or asks which one you meant)
- BOB asks clarifying questions if the request is ambiguous
- BOB does the work — reads the vault, drafts content, runs analysis
- BOB returns the output — some workflows save files during the run (paths are shown); others return a draft for you to save
- You review — check the output and its paths, edit if needed, approve or send back for revision
The vault is both BOB's input (it reads your context) and its output (results are saved back as files). When in doubt, check the paths the run reported — if no path is shown, the draft hasn't been saved yet.
Three ways to invoke BOB¶
1. Natural language — the default¶
Draft a follow-up email to the prospect we met last Tuesday
BOB infers the right skill, reads recent meeting notes, and drafts the email. Use this most of the time — BOB is good at picking skills.
2. Explicit skill name¶
/proposal-writer
Loads exactly that skill and runs its workflow. Use when you know precisely which skill you want, or BOB picked the wrong one. The full index of installed skills lives in SKILLS at the vault root.
3. Playbooks — multi-step recipes¶
Run the new-client-onboarding playbook
A playbook chains several skills into one repeatable process, with checkpoints between steps. Use for recurring multi-step work — onboarding, monthly close, weekly reviews.
Where results are saved¶
BOB respects the vault's structure. Outputs go to predictable places:
| Output type | Saved to |
|---|---|
| Client deliverables | {client.projects} |
| Marketing content | {company.marketing}/{channel}/ |
| Research | {resource.root} |
| Personal notes / inbox | {me.inbox} |
| Meeting notes | the relevant 20-MEETINGS/ folder |
| Tasks | {output.tasks} |
When a workflow saves a file, BOB tells you the path before it writes, and you can override it. A workflow that returns a draft in the conversation saves nothing until you ask BOB to save it.
Habits that get the best results¶
- Point BOB at context. "Using the notes in the Acme folder, draft an upsell proposal" beats "draft an upsell proposal" every time.
- Iterate in the same conversation. "Good draft — shorter, and drop the intro" keeps the thread's context. Don't start over.
- Save what's good, edit the rest yourself. If a draft is 80% there, save it and polish in Obsidian directly — faster than three rounds of tiny tweak requests.
- Keep the vault tidy. BOB reads the vault for context; a stale or messy vault produces noisy answers. Review your inbox folder weekly.
Patterns to avoid¶
| Don't | Why |
|---|---|
| Ask for something no skill covers and accept the answer blindly | Check SKILLS; unsupported requests get generic, unreliable answers |
| Paste secrets or API keys into the chat | Chat content is sent to the AI provider — keep credentials in configuration files |
| Treat output as final without review | BOB drafts; you approve — always review business-critical output |
| Run BOB on out-of-date vault data | If a client folder is six months stale, "current state" advice will be wrong |