Choose the Smallest Native Surface
Choose among a prompt, AGENTS.md, skill, specialist agent, connector, plugin, or schedule.
What you will complete
A surface decision that explains why repository guidance plus one skill fits the chosen job.
Make the operating decision
Use a prompt for one-time context, AGENTS.md for durable repository rules, a skill for a reusable procedure, and a specialist agent only when a distinct role, model, sandbox, tool set, or handoff is required. A connector provides external data or actions. A plugin packages capabilities for distribution. A scheduled task adds recurrence after the manual path works.
Mixed needs can use more than one surface. This course pairs a short AGENTS.md section with a custom skill because the repository needs stable context while the job needs repeatable steps and evals. The agent itself may be the normal file-aware builder following those artifacts. A separate multi-agent coordinator would add task identity, communication, state, and cancellation requirements that the Harbourlight job does not need.
Record rejected alternatives. If a weekly schedule is deferred, say that manual invocation must pass first and name the future scheduling proof. If live CRM access is deferred, retain the local sanitized export. This makes the design smaller without pretending the missing integration exists.
Worked fictional example: Harbourlight Home Services
Harbourlight rejects a one-off prompt because the workflow recurs, and rejects a team because one process can create and check the brief. Repository guidance identifies the source-of-truth files and permission rules. The estimate-exception skill defines the repeatable sequence, schema checks, output, and evals.
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 2
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.
- Compare the job with each native surface.
- Choose the minimum combination and explain it.
- Record why a team, connector, plugin, and schedule are deferred or required.
- State what evidence would justify adding each deferred surface.
Critical gate before continuing
- The chosen surface matches the recurring need.
- No separate role exists without a distinct boundary.
- Manual operation remains possible.
- Deferred features have mechanisms rather than vague wishes.
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 surface decision that explains why repository guidance plus one skill fits the chosen job. 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. Which surface holds durable repository conventions?
Explained answer: AGENTS.md or equivalent project guidance. Project guidance supplies context that should apply across relevant work in the repository.
2. When is a skill appropriate?
Explained answer: When a repeatable procedure or domain workflow should be reusable. A skill carries repeatable decisions, steps, references, and deterministic helpers.
3. When should scheduling be added?
Explained answer: After the exact manual path passes. Manual proof exposes input, permission, output, and stop problems before unattended execution.
Return to Build a Business AI Agent with AGENTS.md, Skills and Evals