---
title: "What Context Should an Agent Read Before It Starts?"
description: "Give an agent the smallest authoritative context set for the task: current facts, decisions, corrections, boundaries, and a clear source order."
canonical: "https://scalewithsearch.com/articles/what-context-should-an-agent-read"
date: "2026-08-17"
modified: "2026-09-19"
---
## Site navigation

- [Scale With Search](https://scalewithsearch.com/)
- Real estate
  - Real estate
    - [Real estate](https://scalewithsearch.com/for/real-estate)
- Work
  - Start here
    - [Send your brief](https://scalewithsearch.com/work#send-your-brief)
    - [Prepare your six-question brief](https://scalewithsearch.com/work#prepare-your-six-question-brief)
  - Build
    - [Site build, content library with SEO, signal desk](https://scalewithsearch.com/work)
- For your business
  - Trades and home services
    - [Auto body and collision shops](https://scalewithsearch.com/for/auto-body-and-collision-shops)
    - [Foundation and home repair contractors](https://scalewithsearch.com/for/foundation-and-home-repair)
    - [Garage door and fencing contractors](https://scalewithsearch.com/for/garage-door-and-fencing-contractors)
    - [HVAC contractors](https://scalewithsearch.com/for/hvac-contractors)
    - [Janitorial and commercial cleaning companies](https://scalewithsearch.com/for/janitorial-and-commercial-cleaning)
    - [Locksmiths](https://scalewithsearch.com/for/locksmiths)
    - [Moving companies](https://scalewithsearch.com/for/moving-companies)
    - [Pest control companies](https://scalewithsearch.com/for/pest-control-companies)
    - [Plumbing and electrical contractors](https://scalewithsearch.com/for/plumbing-and-electrical-contractors)
    - [Restoration and water or fire damage companies](https://scalewithsearch.com/for/restoration-and-water-fire-damage)
    - [Roofing companies](https://scalewithsearch.com/for/roofing-companies)
    - [Towing companies](https://scalewithsearch.com/for/towing-companies)
    - [Tree services and landscaping companies](https://scalewithsearch.com/for/tree-services-and-landscaping)
    - [Solar installers](https://scalewithsearch.com/for/solar-installation)
    - [General contractors](https://scalewithsearch.com/for/general-contractors-and-construction)
    - [Paving, concrete, and flooring contractors](https://scalewithsearch.com/for/paving)
  - Practices and professional services
    - [Bookkeeping and tax practices](https://scalewithsearch.com/for/bookkeeping-and-tax-practices)
    - [Dental practices](https://scalewithsearch.com/for/dental-practices)
    - [Family and criminal defense law firms](https://scalewithsearch.com/for/family-and-criminal-defense-law-firms)
    - [Med spas and aesthetics practices](https://scalewithsearch.com/for/med-spas-and-aesthetics)
    - [Personal injury law firms](https://scalewithsearch.com/for/personal-injury-law-firms)
    - [Veterinary clinics](https://scalewithsearch.com/for/veterinary-clinics)
    - [Gyms and fitness studios](https://scalewithsearch.com/for/fitness)
    - [Therapy and outpatient health practices](https://scalewithsearch.com/for/therapy-and-outpatient-health)
    - [Medical billing companies](https://scalewithsearch.com/for/medical-billing)
    - [Insurance agencies](https://scalewithsearch.com/for/insurance-agencies)
    - [Financial advisors](https://scalewithsearch.com/for/financial-advisors)
    - [Property management companies](https://scalewithsearch.com/for/property-management)
    - [Recruiting and staffing agencies](https://scalewithsearch.com/for/recruiting-and-staffing)
    - [Architects and interior designers](https://scalewithsearch.com/for/architects-and-interior-designers)
    - [Logistics and supply chain companies](https://scalewithsearch.com/for/logistics-and-supply-chain)
  - Agencies, MSPs, and manufacturing
    - [IT and managed service providers](https://scalewithsearch.com/for/it-and-managed-service-providers)
    - [Machine shops and precision manufacturers](https://scalewithsearch.com/for/machine-shops-and-precision-manufacturing)
    - [Marketing agencies and freelancers](https://scalewithsearch.com/for/marketing-agencies-and-freelancers)
    - [SEO agencies and consultants](https://scalewithsearch.com/for/seo-agencies-and-consultants)
    - [Small manufacturers and fabricators](https://scalewithsearch.com/for/small-manufacturers-and-fabricators)
  - Restaurants, shops, studios, and nonprofits
    - [Restaurants and hospitality businesses](https://scalewithsearch.com/for/restaurants-and-hospitality)
    - [Retail stores and ecommerce sellers](https://scalewithsearch.com/for/retail-and-ecommerce)
    - [Photographers, event planners, and travel agents](https://scalewithsearch.com/for/photographers)
    - [Churches and nonprofits](https://scalewithsearch.com/for/churches-and-nonprofits)
  - [All industries](https://scalewithsearch.com/for/)
- Learn
  - For your office
    - [Office job guides](https://scalewithsearch.com/guides/)
    - [Browser calculators](https://scalewithsearch.com/tools/)
  - Start here
    - [How it works](https://scalewithsearch.com/how-it-works)
    - [Free Starter Kit](https://scalewithsearch.com/kit/business-memory-starter-kit.zip)
    - [Synthetic specimen](https://scalewithsearch.com/specimen/working-session-specimen.zip)
  - Guides
    - [The Complete Guide to Business Memory for AI Agents](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide)
    - [The Complete Small-Business Guide to AI Agent Governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business)
    - [The Complete Guide to Leaving Vendor AI Memory](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide)
  - Articles by cluster
    - [Business memory](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide)
    - [Agent governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business)
    - [Migration and ownership](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide)
  - For machines
    - [llms.txt](https://scalewithsearch.com/llms.txt)
    - [llms-full.txt](https://scalewithsearch.com/llms-full.txt)
    - [Machine view](https://scalewithsearch.com/?view=machine)
- Company
  - Evidence
    - [Proof](https://scalewithsearch.com/proof)
  - Company
    - [About](https://scalewithsearch.com/about)

# What Context Should an Agent Read Before It Starts.

An agent receives the entire company archive to draft one renewal note. It finds an old package description, a brainstorming document, and a superseded price. The draft is detailed, fluent, and wrong.

More [context](/articles/stop-re-explaining-your-business-to-ai) did not create more accuracy. It created more unresolved choices.

The right context is the smallest authoritative set that lets the agent complete the named job and recognize when it cannot.

## Context windows are capacity, not judgment

Model documentation describes a context window as the text available to the model during a request. Providers also describe limits and techniques for managing long context. Capacity does not tell a business which records are authoritative or appropriate. [Anthropic: context windows](https://platform.claude.com/docs/en/build-with-claude/context-windows)

A large window can hold more material. It can also hold more stale rules, duplicate facts, unrelated clients, and private information.

Context selection is an information-architecture decision.

## Begin with the task boundary

Ask what the agent must produce.

For a weekly client update, the minimum packet may include:

- current client context;
- work completed during the period;
- active decisions;
- relevant corrections;
- the task brief;
- the approval rule.

It probably does not need every proposal, transcript, invoice, private personnel note, and archived campaign.

For a pricing decision, the packet changes. The agent may need cost inputs, the current offer, prior exceptions, and the owner's decision criteria. It may not need the content calendar.

Start from the output. Work backward to the records required to support it.

## Set a context budget

A context budget is a list of records the job may load by default, records it may retrieve when a condition is met, and records it must never load for that job.

This is not a token budget. It is an authority and relevance budget.

For a weekly account brief, default context might include the current account record, current decisions, active corrections, and the task brief. Conditional context might include an original contract when scope is disputed. Excluded context might include other clients, private personnel notes, payment details, and old drafts.

Write the decision before retrieval:

```text
default:: account-context.md, decisions-active.md, corrections-active.md
conditional:: signed-scope.pdf when scope_conflict is true
excluded:: other-clients/, personnel/, payment-details/, archive/drafts/
maximum_age:: account-context.md 7 days
stop_if:: required record is stale or condition cannot be resolved
```

The budget keeps broad search from becoming the default. It also tells a reviewer why a source was present.

## Use layers

A practical context path has four layers.

The first layer is the job. `task-brief.md` defines the output, reader, acceptance test, and stopping point.

The second layer is current business context. This includes the active account state, offer, scope, dates, and owners.

The third layer is governance. Decisions, corrections, source order, permissions, and exclusions live here.

The fourth layer is evidence. Source documents support claims when the maintained summary is insufficient or contested.

Load the first three by default when they remain small and relevant. Retrieve evidence as needed. Do not flood the run with every historical record. For carrying that source packet between tools, read [Persistent Context Across AI Tools: Files You Keep.](/articles/persistent-context-across-ai-tools)

## Mark authority and freshness

Each operational file should state who owns it and when it was verified. For separating standing instructions from changing business facts, read [CLAUDE.md vs context.md: Rules and Business Records.](/articles/context-file-vs-claude-md-business-project)

```text
owner:: account lead
last_verified:: 2026.08.17
status:: active
supersedes:: 2026.07.01 client summary
```

Freshness is not the same for every fact. A postal address may stay stable. Pipeline status may change daily. A price may change only by owner decision.

The context path should reflect that. If a required file is beyond its review interval, the agent should flag it before relying on it.

Freshness should match the decision. A legal entity name may change rarely but carry high consequence. A work status may change daily but matter only to one report. Do not assign the same review interval to every file.

Use two triggers when they fit: a scheduled review and an event-based review. A weekly status record may need a seven-day check. A scope record may need review whenever an amendment is approved. A personnel-access rule may need review when a person joins, leaves, or changes role.

The agent should not "refresh" the record by inventing an update. It can query an approved system, label the record stale, or prepare a request to the owner.

## Retrieve evidence only when the job needs it

The maintained context should be small enough to inspect. Original evidence can remain in its governed location until a fact is missing or challenged.

For example, `account-context.md` can state the current approved scope and point to the signed agreement. The agent does not need the entire agreement for every routine update. It retrieves the agreement only when the task involves a scope question and the manifest permits that retrieval.

This reduces noise and unnecessary data exposure. It also keeps the distinction between a maintained summary and the evidence that supports it.

## Inspect `context-manifest.md`

Create a manifest for one recurring job:

```text
job:: weekly client update
required:: task-brief.md, client-context.md, decisions.md, corrections.md
conditional:: source-evidence/
excluded:: other-client-folders/, private-notes/, archive/
freshness:: client-context.md verified within 7 days
```

Run three tests.

First, a sufficiency test: can the agent produce the accepted draft from only the required files?

Second, a subtraction test: remove one file. Does the system identify the missing source instead of inventing the answer?

Third, a contamination test: add an old conflicting record. Does the source-order rule prevent it from overriding the current decision?

Record the results in `receipts/context-test.md`.

## Keep client and role boundaries visible

Context must not cross an account boundary because two folders use similar terms.

The manifest should name the account, role, and permitted path. An agent working on Client A should not search Client B by default. A sales agent should not receive private personnel notes because they might help personalize a message.

Access should follow the job, not the theoretical capacity of the model.

Data minimization belongs in the manifest. If a source contains fields the job does not need, retrieve a narrower view or prepare a redacted test record. More access does not make the result more grounded.

The same rule applies to connected tools. A job that reads an approved account summary should not receive broad drive search because a connector supports it. Match access to the records named by the context budget.

Review the manifest when the job, source owner, client, role, provider, or data class changes. A context packet approved for an internal draft does not automatically become approved for an external message or a different model provider.

## Approval and stopping boundary

The agent stops when a required source is missing, stale beyond its rule, contradictory at the same authority level, or outside the permitted path.

It also stops before sending, publishing, purchasing, deleting, deploying, or changing a live record unless the named owner approves the exact action.

The system may prepare a missing-information request or an exact proposed change. It may not fill the gap with a plausible claim.

Use the [context manifest template](/articles/ai-context-manifest-template) to record these boundaries. Then run the [pre-production test plan](/articles/test-an-ai-agent-before-production) against missing, stale, and conflicting sources.

## Context should make refusal easier

A strong context packet does more than help the agent answer. It helps the agent know when the answer is unsupported.


## Questions about What Context Should an Agent Read Before It Starts?

### what user context would actually make ChatGPT better for you?

Start from the output the agent must produce, then work backward to the current facts, decisions, corrections, task brief, and approval rule required to support it. Use a context budget that names default, conditional, and excluded records so broad retrieval does not become the default.

### Best practices (on length) for prompt instructions?

Context selection is an authority and relevance decision, not only a question of token count. Load the smallest relevant packet, retrieve evidence only when the job needs it, and stop when a required source is missing, stale, or contradictory.

### Long ChatGPT chats go bad but starting a new one means losing all your context. How do you actually deal with this?

Use a maintained packet with a task brief, current context, decisions, corrections, source order, and an approval boundary. Test it in a fresh session without remembered context, and treat a missing source as a reason to stop rather than invent an answer.

## Save the visual summary

Begin with the task boundary | Use layers | Retrieve evidence only when the job needs it | Keep client and role boundaries visible

[Download the PNG](/infographics/what-context-should-an-agent-read-1200x1500.png)


## Related: Owned Memory

- [AI Context Manifest Template for One Recurring Business Job](/articles/ai-context-manifest-template)
- [AI Correction Log Template That Changes the Next Run](/articles/ai-correction-log-template)
- [How Corrections Become Part of the System](/articles/corrections-become-system-memory)

----

```text
                  .|########||.                                       .|########||.                                       .|########||.
               |##||.      .||##|.                                 |##||.      .||##|.                                 |##||.      .||##|.
             |#|.              .|#|.                             |#|.              .|#|.                             |#|.              .|#|.
           |#|                    |#|                          |#|                    |#|                          |#|                    |#|
         .#|                        |#.                      .#|                        |#.                      .#|                        |#.
        .#.                          .#|                    .#.                          .#|                    .#.                          .#|
       |#.                            .#|                  |#.                            .#|                  |#.                            .#|
      |#             ......             #|                |#             ......             #|                |#             ......             #|
     .#           ||#########|           #|              .#           ||#########|           #|              .#           ||#########|           #|
    .#.         |######||######|.        .#.            .#.         |######||######|.        .#.            .#.         |######||######|.        .#.
    #.        .##|###|##|#|######|        .#            #.        .##|###|##|#|######|        .#            #.        .##|###|##|#|######|        .#
   ||        |##|#||||||||||||#||#|        ||          ||        |##|#||||||||||||#||#|        ||          ||        |##|#||||||||||||#||#|        ||
   #        |#||||||||||||||||||||#|        #.         #        |#||||||||||||||||||||#|        #.         #        |#||||||||||||||||||||#|        #.
  ||       |#||||||||||||||||||||||#|       ||        ||       |#||||||||||||||||||||||#|       ||        ||       |#||||||||||||||||||||||#|       ||
  #       .#||||||||||||||||||||||||#|       #        #       .#||||||||||||||||||||||||#|       #        #       .#||||||||||||||||||||||||#|       #
 ||   ....|||#||||||##|#|||#|#||##||||....|. ||      ||   ....|||#||||||##|#|||#|#||##||||....|. ||      ||   ....|||#||||||##|#|||#|#||##||||....|. ||
 #.  .  ....|#  ....#|||   ||| .#|||  ....#. .#      #.  .  ....|#  ....#|||   ||| .#|||  ....#. .#      #.  .  ....|#  ....#|||   ||| .#|||  ....#. .#
 #   .  ||||#| .#####||  . .#| .#|||  ||||#   #.     #   .  ||||#| .#####||  . .#| .#|||  ||||#   #.     #   .  ||||#| .#####||  . .#| .#|||  ||||#   #.
.|   |||||  #. |#|||||. ||  #. |#||. .|||||   ||    .|   |||||  #. |#|||||. ||  #. |#||. .|||||   ||    .|   |||||  #. |#|||||. ||  #. |#||. .|||||   ||
||   |....  #. ....|#.      |. ...|. ....||   ||    ||   |....  #. ....|#.      |. ...|. ....||   ||    ||   |....  #. ....|#.      |. ...|. ....||   ||
#.  .||||||##||||||##||####|||||||#||||||#|   .#    #.  .||||||##||||||##||####|||||||#||||||#|   .#    #.  .||||||##||||||##||####|||||||#||||||#|   .#
#    .||####|###########################|.     #    #    .||####|###########################|.     #    #    .||####|###########################|.     #
#      .#||||||.#.|| # |. ..# #| #|||||#|      #    #      .#||||||.#.|| # |. ..# #| #|||||#|      #    #      .#||||||.#.|| # |. ..# #| #|||||#|      #
#      .#|||||| ..  |# ## |#| ...#|#|||#|      #    #      .#|||||| ..  |# ## |#| ...#|#|||#|      #    #      .#|||||| ..  |# ## |#| ...#|#|||#|      #
#      .#|||#|# .# .#| #| ##|.#..#|||||#|      #    #      .#|||#|# .# .#| #| ##|.#..#|||||#|      #    #      .#|||#|# .# .#| #| ##|.#..#|||||#|      #
#      .#||||||############|######|||||#|      #    #      .#||||||############|######|||||#|      #    #      .#||||||############|######|||||#|      #
#   |...||#||||||#||||#|#||||||#|||||#||| |.|  #    #   |...||#||||||#||||#|#||||||#|||||#||| |.|  #    #   |...||#||||||#||||#|#||||||#|||||#||| |.|  #
#. .. ||||# .|||##|.  |#| .|| || .|||#. #|  # .#    #. .. ||||# .|||##|.  |#| .|| || .|||#. #|  # .#    #. .. ||||# .|||##|.  |#| .|| || .|||#. #|  # .#
|| |  ...||  ...##| |  #| .|. |. #####  .. .| ||    || |  ...||  ...##| |  #| .|. |. #####  .. .| ||    || |  ...||  ...##| |  #| .|. |. #####  .. .| ||
|| ||||| || ||||#|  .  |. |. |#. ||||| |#| || ||    || ||||| || ||||#|  .  |. |. |#. ||||| |#| || ||    || ||||| || ||||#|  .  |. |. |#. ||||| |#| || ||
.# |.....#|....|#.||||.|||##.|#|....||.#||.#. #.    .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #.    .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #.
 #.|||||||#############################| ||| .#      #.|||||||#############################| ||| .#      #.|||||||#############################| ||| .#
 ||       ##||#||#||||||#||#||#||#||##.      ||      ||       ##||#||#||||||#||#||#||#||##.      ||      ||       ##||#||#||||||#||#||#||#||##.      ||
  #       .#|||||||||#||#|||||||||||#|       #        #       .#|||||||||#||#|||||||||||#|       #        #       .#|||||||||#||#|||||||||||#|       #
  ||       |#||||||||||||||||||||||#|       ||        ||       |#||||||||||||||||||||||#|       ||        ||       |#||||||||||||||||||||||#|       ||
  .#        |#||||||||||||||||||||#|        #.        .#        |#||||||||||||||||||||#|        #.        .#        |#||||||||||||||||||||#|        #.
   ||        |#||||||||||||||||||#|        ||          ||        |#||||||||||||||||||#|        ||          ||        |#||||||||||||||||||#|        ||
    #.        .######|#|#########|        .#            #.        .######|#|#########|        .#            #.        .######|#|#########|        .#
    .#.         |#####||||#####|         .#.            .#.         |#####||||#####|         .#.            .#.         |#####||||#####|         .#.
     |#           ||########||           #.              |#           ||########||           #.              |#           ||########||           #.
      |#             ......             #|                |#             ......             #|                |#             ......             #|
       |#.                            .#|                  |#.                            .#|                  |#.                            .#|
        |#.                          .#.                    |#.                          .#.                    |#.                          .#.
         .#|                        |#.                      .#|                        |#.                      .#|                        |#.
           |#|                    |#|                          |#|                    |#|                          |#|                    |#|
            .|#|.              .|#|.                            .|#|.              .|#|.                            .|#|.              .|#|.
               |##||.      .||##|                                  |##||.      .||##|                                  |##||.      .||##|
                 .||########||.                                      .||########||.                                      .||########||.

Scale With Search  2026  [scalewithsearch.com](https://scalewithsearch.com)
```