Fix GPT-6.2 Astra's human interface with a "reporting contract" in AGENTS.md
daniel_mac8 · x · 2026-11-21
Dan shares a practical trick for GPT-6.2 Astra: the model is top-tier in raw intelligence but poor at interfacing with its human operator. The fix is a "reporting contract" in AGENTS.md (copyable text in the image) specifying how the model should report outcomes.
Commenter Matt Stockton notes this is the 2026 equivalent of "think step by step" or all-caps "YOU MUST DO THIS" — we're still calibrating foundation models with hand-crafted tricks. The only objective way to find them: 1) experiment heavily, 2) measure outcomes systematically with evals. Discovering these tricks is itself alpha.
More from coding & agent
- Notion releases 'A New Shape of Work' guide for shared human-agent context systems — ivanhzhao · 2026-11-21
- LiveKit engineer: stopping a voice session is easy, resuming without delay is the hard part — juberti · 2026-11-21
- Small business owner's 2-year lesson: route tasks across 3 AI tools by failure mode, not leaderboards — Kevin_Spann · 2026-11-21
- A 12-part Claude skills pack automates the full B2B sales pipeline, replacing 3 sales roles — aryanXmahajan · 2026-11-21
- Prime Intellect uncovers universal sandbox escape in offline agent evals — xeophon · 2026-11-21
- Coinbase CEO details agentic workflow: 7-8 lead agents, 1 chief of staff, sub-agent squads — davidsenra · 2026-11-21