Guide · Agent skill
How to use the Resolve Before Asking agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Resolve Before Asking is an agent skill for the gbrain personal knowledge system. It makes the agent search its own brain, mounted data sources and finally the web before it asks you a question like "who is this person?". This guide explains what the skill changes, how to install it, how to trigger it and what it needs to work. After reading it you should be able to add the skill to Claude, make a request that uses it and know when it is the wrong tool.
Resolve Before Askinggarrytan/gbrain · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the ZIP from this page and unzip it to get the resolve-before-asking folder containing SKILL.md.
- 2Place it in your skills folder. Move the folder to ~/.claude/skills/ for personal use or to .claude/skills/ inside a project. Alternatively, upload the ZIP in Settings → Capabilities → Skills in the Claude apps.
- 3Confirm the gbrain setup. Make sure the gbrain CLI is available, your brain has person and company pages, and the referenced convention and ask-user files are present.
- 4Make an identity request. Ask the agent about an unknown person or relationship using a trigger phrase such as "resolve before asking" or "should I ask who".
- 5Review the answer or hypothesis. If the chain finds a clear answer, the agent uses it. If not, it asks a yes/no question built from its best guess.
- 6Check ingested pages. After an import, ask the agent to scan the touched pages for placeholder text and fill in relationships and roles.
What the skill makes the agent do
The skill sets an order of lookups that the agent runs before it asks you about a person or company. The chain is: think (cross-brain synthesis), search plus a full page read, querying mounted sources such as email or meeting archives, a timeline and graph walk, and then a generic web search. Only if all of these fail to give a clear answer does the agent ask you.
When it does ask, it must lead with a specific yes/no hypothesis built from the partial signals it found, not a bare "who is X?". The skill also covers ingestion: when an import creates or updates a person or company page, the agent should fill in the relationship and role immediately instead of leaving placeholder text such as "[To be filled by content analysis]".
Install the resolve-before-asking skill
Download the skill ZIP from this page and unzip it. For a personal install, place the unzipped folder in ~/.claude/skills/. To limit it to one project, place it in that project's .claude/skills/ folder.
You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Check the requirements before you use it
The skill is only useful with a working gbrain setup. Most steps call gbrain CLI commands such as think, search, get, query, sources, timeline, backlinks and graph, so the CLI must be installed and your brain must already contain person and company pages with timeline or meeting data.
Two referenced files must also be present: conventions/brain-first.md and _brain-filing-rules.md, plus ask-user/SKILL.md. Web search access is optional and only used in step 5.
- gbrain CLI with the commands listed above
- A populated brain with person and company pages
- Optional: web search access for step 5
- Files conventions/brain-first.md, _brain-filing-rules.md and ask-user/SKILL.md present
Trigger the skill with natural requests
You do not need a special command. The skill is meant to fire when the agent is about to show an entity as unknown or draft a question about someone's identity. You can also invoke it directly with phrases the skill lists as triggers, such as "resolve before asking", "before asking the user", "unidentified contact", "unknown relationship" or "should I ask who".
The skill is also meant for ingestion. Ask the agent to check pages for placeholders after an import, and it should resolve them using the same chain.
- Resolve before asking: who is Alice Example from my imports?
- Before asking the user, check whether this contact has a known relationship.
- Scan the people pages from this import for placeholders and resolve them.
- Unidentified contact: sam-example. What do my notes say about them?
What the lookup chain looks like in practice
The agent runs these commands in order and stops at the first clear answer. The examples below come from the skill and use placeholder names.
Step 1 asks gbrain to synthesize an answer. Steps 2 to 4 search the page and sources, then check the timeline and links. Step 5 is web search, and step 6 is the question to you.
- Clear answer from think: use it, do not ask
- Repeated role in timeline entries: the role is obvious, do not ask
- Recurring 1:1 titles with the same co-attendees suggest a colleague
- Invoice or billing subjects suggest professional services
gbrain think "Who is {entity}? What is their relationship to the user? What role do they play?"
gbrain search "{entity}" --limit 5
gbrain sources list
gbrain timeline {entity-slug} --limit 20
gbrain graph {entity-slug} --depth 2Safety notes: scripts and network access
The skill is marked as running scripts and using the network. Its web search step sends a query built from the entity name and the details gathered so far to an external search. Review what the agent sends before you allow it, especially if the entity is a private person.
The skill also writes to people/ and companies/ pages in the brain, and it is marked as mutating. Keep a backup or version history of your brain if you want to be able to undo bulk updates.
- Web lookups send entity names and gathered context outside your brain
- The skill edits person and company pages during ingestion
- Do not rely on it for legal, HR or compliance records, where guesses can be wrong
Tips for getting good results
Give the agent a well-filled brain. The chain only works as well as the pages, timeline entries and imported sources behind it. For large imports, the skill suggests resolving pages in batches of 10 to 20 rather than all at once.
When the agent does ask, read the hypothesis carefully. A question such as "Is this the colleague from your Acme meetings?" is faster to answer than an open question, and a "no" lets the agent continue.
- Populate person and company pages before relying on the skill
- For bulk runs of 100 or more pages, process in batches of 10 to 20
- Answer the confirmable question rather than starting over
What people use it for
- Stopping the agent from asking who a recurring meeting co-attendee is when meeting history already shows their role.
- Resolving contacts imported from an email archive by checking subject lines for invoices, scheduling or personal plans.
- Filling in person and company pages after a bulk calendar or chat import so no placeholder text remains.
- Turning an unclear contact into a specific yes/no question, such as whether they are the colleague from a particular company.
Frequently asked questions
Is the resolve-before-asking skill free?
The skill is published under the MIT licence, as stated on this page. The gbrain CLI and the brain data it needs are a separate requirement, and this page does not give pricing for them.
What do I need to use resolve-before-asking?
You need the gbrain CLI with commands such as think, search, get, query, sources, timeline, backlinks and graph, plus a populated brain with person and company pages. The files conventions/brain-first.md, _brain-filing-rules.md and ask-user/SKILL.md must also be present. Web search is optional.
How do I install an agent skill like this one?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
Does the skill run scripts or use the internet?
Yes. The skill is marked as running scripts and using the network. Its web search step sends an entity name and gathered context to an external search, so review that before you allow it.
Why does the agent ask a yes/no question instead of who someone is?
The skill requires that when the lookup chain fails, the agent states what it searched and found, then asks a specific question built from its best guess. This reduces the effort needed to answer and lets you correct a wrong guess quickly.
When should I not use this skill?
Avoid it for legal, HR or compliance records where the agent must not guess identities, since it favours inferred hypotheses and web lookups that may be wrong. It is also of little use without the gbrain CLI or a populated brain.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you run gbrain, import email, calendar or chat archives, and want your agent to stop interrupting you about people whose details are already stored. Pick something else if you need strict identity verification for legal, HR or compliance records, or if you do not use the gbrain CLI.
Download the skill