Project Setup

What it measures

Does your briefing file describe what's unique about your project, rather than general instructions the model already knows?

The four diagnostic questions

1. 01Your project instructions are scoped to proprietary context only — not generic facts the model already knows.
2. 02Your briefing file is updated after every significant decision or scope change. It reflects current state.
3. 03The briefing is concise. A short accurate briefing is preferred over a long stale one.
4. 04You have explicit prohibitions in your briefing: what the agent must not touch, change, or assume.

What good looks like

Strong setup discipline. Ensure the briefing is versioned alongside your work, not manually maintained in isolation.

Common failure mode

Strip your briefing to proprietary context only and update it to current state. A short accurate briefing outperforms a long stale one every time.

Critique & references

Editor's notes — to be expanded

The 500-word briefing mandate is validated by PromptSet (114K prompt variants), which shows underspecified prompts increase output variance and lower model confidence (logits). With Sonnet 3.7+ enforcing hard validation errors on missing context, briefing hygiene shifts from best-practice to functional requirement.

• → PromptSet — 114K prompt variants (arXiv 2402.16785)
• → Anthropic — Prompt engineering overview