The first line carries the load.
The i-have-adhd project is a real, MIT-licensed coding-agent skill. Its rules put the next action first, split multi-step work into numbered actions, restate progress, suppress tangents, and end with one concrete move.[1]
The useful idea is larger than one diagnosis or one plugin. Agent output is an interface. The reader should not need to mine a transcript to find the command, file, decision, or failure.
A correct answer can still have a bad clutch.
Do not confuse short with thin.
Michael Lynch's software-blogging checklist asks authors to show the payoff in the title and first three sentences. It also says links should extend an explanation rather than replace it. The same rules fit coding-agent output. Lead with the result. Keep enough context on the page to use it.[2]
A terse answer that drops the failing test, changed path, or safety boundary is not efficient. It pushes work back to the reader. Good compression keeps the action, state, evidence, and next move. It removes throat-clearing, repeated plans, and unrelated cleanup.
Use disclosure, not a text wall.
Native HTML already has a control for optional depth. The <details> element creates a disclosure widget with a <summary> label. Browsers expose the open and closed state without a custom script.[3]
That gives an answer two layers. The visible layer holds the decision and immediate action. The disclosure holds logs, alternatives, citations, and long explanations. Keyboard users keep a normal control. Readers who need the proof can open it where it matters.
Adopt a four-part answer contract.
RESULT=[WHAT CHANGED OR FAILED] ACTION=[ONE COMMAND, PATH, OR DECISION] EVIDENCE=[TEST, STATUS, OR SOURCE] DEPTH=[INLINE / DISCLOSURE / REQUESTED]
Use the smallest gear that completes the task. Move up when the reader asks for an explanation, the action is destructive, uncertainty changes the decision, or the proof cannot fit beside the result.