Worked example · Engineering teams and technical writers · Reviewed October 27, 2026

We loaded the public skill below through 9octopus and applied it to this sample input. The output is shown so you can inspect the result and repeat the task. Your agent’s output may differ.

What you need

A connected agent, the decision notes, and any existing ADR naming convention. Reading and drafting are enough for this example; writing into a repository requires file access.

Choose your agent’s setup guide, connect it to 9octopus, and finish OAuth sign-in before trying the prompt.

Skill used

documentation-and-adrs — source revision used

The source link pins the instructions used for this example. Catalog availability can change; find currently available skills for this task.

Example input

Sample project: a public skill catalog. Proposed decision: fetch 24 skills initially, then use an explicit Load more button with a cursor. Considered: load everything; infinite scrolling. Constraints: the catalog can grow, browsing position should be deliberate, and visitors need a clear loading state. No benchmark has been run.

Try this prompt

Find a public documentation skill, load it, and confirm its exact version. Turn these notes into a proposed ADR with context, decision, alternatives, and consequences. Check for an existing ADR convention first. Preserve the 24-item batch size and uncertainty about performance; do not invent benchmark results.

Public task recommendations use OpenAI for semantic matching. Keep secrets and private repository contents out of discovery queries; provide the working material directly to your agent under your normal access controls.

Ask your agent to search the public catalog, check the selected skill’s source and prerequisites, and confirm its version before continuing. If the source revision above is unavailable, review the suggested alternative before applying it. A new search may return different matches.

How we worked through it

  1. Check the destination project for an existing ADR format. This standalone example has no existing ADR sequence.
  2. Separate the decision from its rationale, and explain why each alternative was not selected for the sample project.
  3. Compare the draft with the notes. Preserve the proposed status and mark performance as unmeasured.

Example output

# Proposed decision: explicit catalog pagination

## Status
Proposed — standalone example, not an adopted project policy.

## Context
The public catalog can grow. Visitors need deliberate browsing and visible loading feedback.

## Decision
Fetch 24 skills initially. Offer an explicit Load more button and pass the returned cursor for the next batch.

## Alternatives
Load everything: simpler interaction, but the initial response grows with the catalog.
Infinite scrolling: fewer clicks, but loading happens as visitors move through the page.

## Consequences
Visitors control when another batch loads. The interface must handle loading, failures, and the end of the list. Browsing requires an extra action per batch.

## Open question
No performance benchmark has been run. Measure before claiming a speed improvement.
Download example output

What we checked

  • The draft preserves the 24-item initial batch, explicit action, and cursor-based continuation.
  • Both supplied alternatives and the extra browsing action remain visible.
  • Status is Proposed and the absence of performance measurements is explicit.

What this example does not establish

This is a worked example from supplied notes, not a record of the historical decision behind 9octopus. A project owner must review the decision and fit it into the project’s existing ADR convention before adoption.

Explore another worked example or connect your agent to try your own task.