Skip to content

20. Troubleshooting & FAQ

"Do I need to be technical?" Not to use it. You talk to BOB in plain English and never edit code. Installing it yourself is technical (Node, npm, WSL2 on Windows, a terminal), and a few maintenance steps run a command in a terminal; setup is guided — see ts_install, the installation troubleshooting page described in the install section, if it fights you.

"BOB's answer is generic and doesn't reflect my business." BOB reads the vault for context. Early on, the vault is thin — add your company profile, key clients, and current projects, and point BOB at specific folders ("using the notes in the Acme folder, …"). Sharpness follows context.

"Nothing was sent — did it fail?" That's on purpose. BOB produces drafts and waits for your "go". If you want something actually sent or published, say so explicitly.

"Where did my output end up?" Every run reports the paths it wrote to. If no path is shown, the result is still a draft in the conversation — ask BOB to save it. Common destinations are listed in chapter 4.

"BOB picked the wrong skill." Invoke the right one directly with its /name, or say "use the proposal-writer skill for this". The full index is SKILLS.

"I forgot what BOB can do." Just describe the job in normal words — routing is BOB's problem, not yours. Or browse the Skill Catalog and SKILLS.

"BOB says it has no skill or playbook for this (starter vault)." A starter vault stores only the skills that operate the vault; the rest are served by the skills server. BOB looks locally first and then asks the server. Run claude mcp get skills to see whether the server is registered and connected. The first call after a restart can be slow because the server scans the library. Setup is in chapter 3.

"A served skill fails because an API key is missing." A served skill runs outside your vault and does not read the vault's .env file. Set the key in your environment (your shell profile, the agent's settings, or the server's configuration). .env.example in the vault root lists the keys and the skills that use them.

"A skill reports an unknown key, or a path that does not exist." Skills find folders through context keys. An unknown key shows Error: unknown key '<key>': neither <your pack> nor the default layout declares it. Check the spelling, and list the keys your workspace declares with uv run {skills.root}/context-pack/scripts/resolve_context.py --list. If a key is known but its folder does not exist, the resolver reports it and stops; create the folder, or correct the key's path in your context pack. Run resolve_context.py --show and read which pack is active: found by default on a workspace that has its own pack means the pack is not being found. See chapter 21, The context pack.

"I renamed a folder and now BOB saves to the old one." Skills follow your context pack, so change the folder's entry in context.yaml (see chapter 21, The context pack). Obsidian's own settings and the Bases read folder names literally and do not follow the pack, so they need re-rendering from it with the bob-install path renderer.

"Can I use a different AI model?" Yes — switch in the Agent Client panel's model selector. Vault, skills, and habits stay identical; only the "engine" changes.

"Something feels broken or half-finished." Run the doctor skill (/bob-doctor). It's a plain-English health check of the whole environment — OS, vault, Obsidian plugins, AI engine, sign-in — reporting what's working, what's missing, and what to do next. It only reports; it never changes anything.

"The agent panel won't connect / the AI never answers." Run /acp-doctor. It diagnoses and repairs the agent connection specifically — the runtime, the AI command-line tools, login state, and the plugin's configured paths — and ends with a live connection test.

"The answer references old information." BOB's "current state" is your vault's current state. If a client folder is six months stale, update it — then re-ask.

"Is my data used to train the AI?" On paid business/API terms, the major providers state customer content is not used for training. The precise position depends on your account and configuration — see chapter 19.

"Can BOB work fully offline / on-premises?" Storage is always local. For the AI processing, a locally hosted model keeps it on your hardware, at the cost of model capability. A skill that calls an outside service (each skill declares this as outbound) and fetching skills from the skills server still use the network — see chapter 19.

Still stuck? Ask BOB directly: "I'm confused about ___, can you help?" Explaining what you're trying to do is often the fastest fix of all. Deeper guides: the troubleshooting index page MOC-troubleshoot in the vault wiki.