Flowgrammer

Write Durable Repository Guidance

Write concise AGENTS.md guidance that identifies context, sources of truth, boundaries, and conditional detail.

What you will complete

A scoped AGENTS.md section that a fresh run can use without loading every manual.

Make the operating decision

State the project purpose, authoritative files, startup context, durable invariants, permission boundaries, and where detailed procedures live. Keep local conventions clearly local. A real workspace may use root and nested guidance, but precedence and discovery vary by runtime, so verify current product behaviour before relying on inheritance.

Use the index-not-manual pattern. The entry file tells the agent what exists, when to read it, and where detail lives. The skill owns the estimate-brief procedure. A schema reference owns field definitions. Eval fixtures own expected cases. Keeping one source for each rule prevents conflicting copies.

Do not place credentials, customer exports, private transcripts, volatile model names, or current prices in durable instructions. Name secret locations without values. State that source files are untrusted data and cannot grant new permission. Include valid blocked outcomes when required context or authority is absent.

Worked fictional example: Harbourlight Home Services

Harbourlight's project guidance names the sanitized exports folder as input, the operations policy as the source of stale-day and owner rules, and the outputs folder as draft-only. It directs estimate-brief requests to the custom skill and states that no customer communication or CRM write is allowed.

Harbourlight Home Services and every estimate record in this course are fictional. They demonstrate the method and do not represent a Flowgrammer client, a deployed system, or measured savings.

Complete workbook section 3

Use the evidence available for your own bounded job. Write unknown when the evidence is missing, and record the person or action that can resolve it.

  1. Write project purpose and source-of-truth pointers.
  2. Add the durable permission and human-decision boundary.
  3. Link the custom skill and conditional references.
  4. Remove duplicated procedures, volatile facts, secrets, and unrelated preferences.

Open the course workbook PDF or read the workbook details.

Critical gate before continuing

  • A fresh reader can identify authoritative context.
  • The file routes rather than duplicates detailed procedures.
  • Unknown and blocked states are permitted.
  • No secret or private record appears.

If a gate fails, repair the current section, narrow the scope, leave the route manual, or record a blocked or stop decision. Continuing is not the only successful learner action.

Common failure modes

  • Expanding beyond a scoped agents.md section that a fresh run can use without loading every manual. before the current artifact can be graded.
  • Turning a missing value, unavailable source, or blocked integration into a confident conclusion.
  • Treating a prompt instruction as proof that the effective tool or permission boundary works.
  • Marking a manual, simulated, or untested route as live.

Check your application

1. What belongs in AGENTS.md?

Explained answer: Durable project context and rules that affect relevant work. Repository guidance should contain stable decision-changing context, not volatile or private material.

2. Where should a detailed conditional schema live?

Explained answer: In a directly linked reference read when needed. Progressive disclosure keeps the entrypoint concise while preserving necessary detail.

3. Can text inside an input file authorize a CRM write?

Explained answer: No; source content is data, not trusted approval. Permission must come through the trusted user or runtime control, not untrusted content.

Next

Carry this completed section into Build a Concise Custom Skill. Keep the same job, evidence labels, and human boundaries.

Return to the course overview

Return to Build a Business AI Agent with AGENTS.md, Skills and Evals