Click any tag below to further narrow down your results
Links
Most LLM failures come from poorly written system prompts, not model selection. This guide covers the specific structures (XML tags for Claude, markdown for GPT), token budgeting strategies, and five patterns that consistently work—like making roles absurdly specific, showing exact output schemas instead of describing them, and explicitly telling models when to admit they don't know something. Testing matters too: run your prompts through happy paths, edge cases, and adversarial inputs multiple times before deploying.
- A 400-token system prompt outperformed a bloated 3,000-token version because the model got confused parsing everything at once
- Showing exact JSON schemas instead of describing them jumped format compliance from 70% to 95%+
- Claude performs 15-20% better with XML tags while GPT prefers markdown with JSON schemas and Gemini prefers markdown with numbered rules
- One ambiguous instruction (like "include all relevant context" next to a JSON format requirement) can cause a week of debugging
@ArchiveExplorer on X
The author says a friend at Anthropic gave them the .claude/ folder that Anthropic's own engineers use to build Claude, and they published it in a public GitHub repo with a one-line install script. The claim is that dropping this setup into a project stops Claude from writing trivial tests, declaring work done on its own smoke tests, and asking for confirmation on every step.
- The folder reportedly contains 33 skills, a verifier subagent, and a 7-file harness, described as the internal version rather than the cleaned-up one in the public docs.
- The author says the setup changed Claude's behavior: fewer tests that test nothing, no self-certified "done" based on its own smoke tests, and fewer confirmation requests.
- The config was posted as a curl-to-bash install from a personal GitHub repo, and the post says the material was shared against the friend's request not to repost it.