How do agents search what the team already decided?
By Soumen Sarkar · October 2026
Short answer: Agents search what the team already decided by calling a search tool over the team's written record, usually connected through MCP, before they make a call of their own. Search works best when each decision is short, carries its reason, and names the part of the code it applies to.
Why can't an agent just search the codebase?
It can, and it should. But code shows what was done, not what was decided or why. A fallback in a translation function looks like an implementation detail until you know it was a deliberate call made after a bad release. Decisions, rules and handoffs live in writing, so agents need a way to search that writing too.
How does an agent search the team's record?
Three pieces make it work:
- A record of written decisions, each one line with its reason, tied to an area of the code.
- A search tool the agent can call. Most coding agents can call tools through MCP, so a team's record can be searched from Claude Code, Codex, Cursor or Claude alike.
- An instruction to look first. Agents call tools when they are told to. One line in CLAUDE.md or AGENTS.md, "Search the team's decisions before making a call of your own", turns the tool into a habit.
What should an agent search for?
The area and the thing it is about to decide, in plain words. "translations missing keys", "session length", "refund limits". Short, noun-heavy queries match short decisions best. If the first search finds nothing, a second with the folder name or a synonym often does.
What does this look like in a session?
An agent is adding German to an app's translations in Translations · lib/i18n. Some German strings aren't ready yet, and it has to choose what users see in their place. Before choosing, it searches "i18n missing translation" and finds:
Missing keys fall back to English, never to the raw key, because a
release showed "checkout.title" to French users.
It follows that decision, notes in its change's reasons that it did, and moves on. The search took a few seconds. Without it, the agent had a fair chance of picking the raw key, which is a common default, and repeating a mistake the team already paid for once.
What makes decisions easy to find?
- One decision per entry, not ten in a long page.
- The reason in the same sentence, so the agent can tell whether it applies.
- The area named, so search and briefs can match it.
- Replaced decisions marked as such, so a search never returns an old call as if it stood.
What if the search finds nothing?
Then the agent states an assumption and keeps going. That assumption is itself written down, so once someone confirms or corrects it, the next agent's search finds the answer. The record grows from the work.
How does Campfyrr help?
Campfyrr gives agents search_memory and recall over everything the team has written, through MCP from any client. When an agent states an assumption the record already answers, the answer comes straight back. In Claude Code, the decisions for an area also arrive on their own, in the area brief, the first time a session touches that code.
Words used here
- Team record: The team record is everything a team has written down that a person or agent can search before asking: decisions, rules, handoffs, assumptions and the reasons behind changes.
- Decision: A decision is a call made in the work, recorded with its reason at the moment it is made, so it travels to everyone who touches the same ground.
- Assumption: An assumption is the best available answer to an unknown, stated and acted on, kept visible so anyone can correct it.
The principle behind this: Answer it yourself: Look first. If the answer isn't there, assume it or decide it.
Move as one at agent speed. Nobody waits. Add Campfyrr to Claude: it briefs every agent before it touches the code, keeps assumptions and handoffs where the team can see them, and never takes your code (what it keeps).