An agentic prompt has three readers — you, your team, your agents — and it is good only when it works for all three.

Principle

Stakeholder Trifecta

The three readers

One prompt. Three audiences. Non-overlapping needs.

👨‍💻

You

writer + maintainer

• re-orient fast weeks later
• delta-edit without breaking it
• skim with zero cognitive tax

If skipped — you rewrite it every quarter because you can't remember what your own cryptic sections meant.

👥

Your team

extender + auditor

• understand without your conversation
• modify safely — no hidden contracts
• compare across the codebase

If skipped — the prompt library becomes a single-author asset. Every new hire needs a verbal tour from the author.

🤖

Your agents

parser + executor

• unambiguous workflow steps
• resolvable variables ($1, {VAR})
• dense operands — paths, verbs, must-markers

If skipped — works when you read it back, fails with sub-agents or cold sessions. "It worked yesterday."

Optimizing for one stakeholder at the expense of the others is the dominant cause of prompt churn.

What violation looks like

Four shapes of trifecta-hostile prompts.

failure · 01

Personal shorthand

Cryptic variable names, inside-joke sections, assumes conversation context no teammate has. Runs fine — for one human.

hostile to: team

failure · 02

Agent-only instructions

Reads like machine code. Humans can't audit before execution. No Purpose section — no one can tell what it does without reading the full workflow.

hostile to: you + team

failure · 03

Diary-tone prose

Directed at a human reader, not the agent. Sub-agents can't parse intent. Works in interactive sessions, drifts in cold ones.

hostile to: agents

failure · 04

Consistency debt

Purpose here. Goal there. Objective elsewhere. Each prompt looks fine alone; the aggregate is hostile. The worst violation because it's invisible.

hostile to: all three

A prompt that only works for its author will be rewritten within weeks.

The completion criterion

Three passes before commit. If any fails, it isn't done.

read the prompt three times — once as each reader

YOU

Could I reconstruct what this does from just the top two sections?

re-orient

TEAM

Could a teammate execute or extend this without asking me questions?

auditable

AGENTS

Does every workflow step have explicit operands — paths, variables, verbs — the agent can bind to?

executable

The trifecta IS the completion criterion — not a nice-to-have review.

If any pass fails, the prompt is not done. Ship it anyway and you pay the trifecta tax in rewrites.

The biggest lever

30 bespoke prompts vs 30 prompts with one shape.

Cleverly bespokeN = 30

/deployGoalStepsOutput

/migrateAboutFlowReturns

/reviewObjectiveProcessResult

/seedPurposeTasksReport

/auditSummaryWorkAnswer

Each prompt costs anew. No muscle memory. Agents reparse shape every time. Humans scan top-to-bottom hunting for familiar anchors.

One shape, repeatedN = 30

/deployPurposeVarsFlowReport

/migratePurposeVarsFlowReport

/reviewPurposeVarsFlowReport

/seedPurposeVarsFlowReport

/auditPurposeVarsFlowReport

Reading the Nth one is free. You know where Purpose lives. Team scans predictably. Agents parse reliably, converge on intent faster.

Consistency beats cleverness — it is the single operational lever that compounds benefit across all three stakeholders.

Why this is the meta-principle

All three must converge on the same mental referent.

✗ Divergent names

PurposeGoalObjectiveAboutSummary

You skim for Purpose. Team grep Goal. Agent looks for Objective. All three miss. The prompt breaks silently across readers.

✓ Convergent name

Purpose

Pick one. Use it everywhere. You, team, and agent all land on the same section. The prompt becomes a public interface — and the convention a contract.

Breaking changes to section names are breaking changes to the team/agent contract — treat them like API changes.

The discipline

A prompt is an artifact, not a note — it is read by three stakeholders, and the weakest read sets the ceiling.

Write every prompt for all three readers. Pick a section vocabulary once and hold it. The trifecta is the completion criterion — before you commit, read the prompt three times, once as each reader.