AI Context JSON: A Reusable Project Brief Schema

AI context JSON is a structured project record that tells an assistant what the work is, which rules apply, and where to start next. Use the small schema below to keep those fields consistent, then refresh their contents from the actual project record before each session. fenbs can hold the authoritative tasks and decisions that your exported summary describes.

JSON is a text format with named fields, lists, and values. The format makes information easier for software to check, but it does not make the information true or cause an assistant to read it automatically. The workflow still needs a clear source, a deliberate read, and human review of decisions.

What should a small project context record contain?

Start with information that changes the next action: project identity, the immediate goal, approved rules, the next step, the last update time, and source locations. Keep those fields short enough to inspect together. The point is to make an ordinary handoff predictable, not to reproduce an entire workspace in one file.

• Project: the unambiguous name or identifier that distinguishes this work from similar jobs.
• Goal: the outcome of the current task, rather than a broad description of the organization.
• Rules: approved instructions that apply to this work; suggestions belong elsewhere.
• Next step: a concrete action, including an unresolved prerequisite when work must wait.
• Updated at: when the contents were last reviewed or changed, expressed consistently.
• Sources: locations someone can inspect to confirm the facts and decisions summarized here.

Our context engineering collection covers selecting useful information for a model. This template owns the smaller job of representing a project summary consistently. You can use it with a local file or a deliberate copy into an app that accepts text, provided you confirm what the app actually receives.

Which JSON schema can you copy and validate?

A schema describes the expected structure of a record. This one requires six fields and rejects unexpected field names, which helps catch accidental spelling changes. It is intentionally small: there are no credentials, customer records, or personal profiles. Use an empty list when no sources or rules have been recorded; do not invent entries to fill the template.

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "title": "ProjectContext",
  "type": "object",
  "additionalProperties": false,
  "required": ["project", "goal", "rules", "next_step", "updated_at", "sources"],
  "properties": {
    "project": { "type": "string", "minLength": 1 },
    "goal": { "type": "string", "minLength": 1 },
    "rules": { "type": "array", "items": { "type": "string" } },
    "next_step": { "type": "string", "minLength": 1 },
    "updated_at": { "type": "string", "minLength": 1 },
    "sources": { "type": "array", "items": { "type": "string" } }
  }
}

The JSON Schema getting started guide↗ (opens in a new tab) explains object properties, required fields, and validation. A parser first checks whether a document is valid JSON. A schema validator then checks whether that document follows your selected structure. Neither check verifies that a rule was approved or that a linked source supports the summary.

This schema treats dates and sources as nonempty text or lists of text, without enforcing a timestamp or URL format. That keeps the example easy to adapt to internal references as well as web addresses. Add stricter validation in your own application if it requires a specific timestamp format or only allows certain source locations.

What does a completed JSON example look like?

Say a small design studio is revising a contact form. The following fictional record identifies the approved constraint and an outstanding check. Its source address is a placeholder; replace it with an accessible project reference rather than assuming the sample page contains evidence.

{
  "project": "Studio contact form",
  "goal": "Prepare the revised form for review.",
  "rules": [
    "Collect email only; do not request a phone number.",
    "A person must approve changes to collected information."
  ],
  "next_step": "Test message delivery; the form is built but delivery is unverified.",
  "updated_at": "2026-10-21T14:00:00Z",
  "sources": ["https://example.com/approved-form-requirements"]
}

Notice the difference between the goal and the next step. “Prepare for review” describes the outcome. “Test message delivery” tells the assistant where to begin. The qualification about an unverified result prevents a new session from interpreting “built” as “finished.” If approval is missing, write that dependency directly into the next step.

Keep longer evidence outside the record and link to it. A task reference, decision reference, or document location is more useful than a copied conversation with unclear authorship. The guide to AI context files compares ways of delivering instructions; a JSON record still needs a known loading method in the app you choose.

How does the same record look in AI context YAML?

YAML is another structured text format, often chosen when people edit configuration files by hand. Here is the same example with the same field names and values. Choose one authoritative representation for editing; maintaining separate JSON and YAML copies creates a second opportunity for the next step to disagree.

project: "Studio contact form"
goal: "Prepare the revised form for review."
rules:
  - "Collect email only; do not request a phone number."
  - "A person must approve changes to collected information."
next_step: "Test message delivery; the form is built but delivery is unverified."
updated_at: "2026-10-21T14:00:00Z"
sources:
  - "https://example.com/approved-form-requirements"

The YAML specification↗ (opens in a new tab) defines mappings, sequences, quoting, and indentation. Use spaces consistently and quote text that could be interpreted as a special value. The quoted timestamp in this example stays a string when parsed, which makes comparing it with the JSON version straightforward.

Do not assume every app automatically discovers a file because its extension is familiar. Confirm the filename and loading behavior your app supports, or provide the selected content explicitly. If the record and its references consume too much working space, AI context compression addresses shortening the material without changing the purpose of this schema.

Which update rules prevent a structured record from going stale?

Update the record after a material change, not merely after a certain number of messages. A changed scope decision, finished check, blocked task, or new next step should trigger a review. Change the timestamp when you review the contents; changing it automatically while retaining unchecked text would imply freshness you have not established.

1. Read the authoritative task and decision records before editing the summary.
2. Replace outdated current-state text while preserving history in the system that owns the work.
3. Parse the edited document, validate it against the schema, and inspect the actual values.
4. Ask the next assistant to identify the goal, applicable rules, and unverified next step before acting.

A structured mistake remains a mistake. Avoid promoting a suggestion into the rules list just because it appeared in an assistant’s answer. People approve decisions; the summary records those choices. The guide to AI context rot explains why unchecked material can weaken a session even when it fits. fenbs Decisions and rules keeps approved choices separate from the short packet you prepare for a session.

How to do this with fenbs

1. Create a free account, then add the real project your sample record will describe.
2. Record the current tasks, the decisions people approved, and Where we left off. Keep source links with the relevant work.
3. Connect a supported assistant using Integrations. Ask it to read the project context and draft the six fields from those records.
4. Review the draft, validate it in your own workflow, and save progress after work so the next summary reflects the latest task state.

fenbs does not provide automatic JSON or YAML import or export for this template. If you only need to validate configuration files inside an application, a local parser and schema validator may be enough. fenbs helps when the underlying tasks, decisions, and handoffs need a shared home across sessions.