14 techniques, live from D1
Techniques
Real prompting techniques and guardrails, each citing the exact doc it came from. Filter by category:
| title | category | applies to | guidance |
|---|---|---|---|
| The reference is organized model-specific, then general, then migration tec_best_practices_structure |
general | all |
Anthropic's own prompting-best-practices reference for Claude's latest models (Fable 5, Mythos 5, Opus 4.8/4.7/4.6, Sonnet 5/4.6, Haiku 4.5) is organized in three parts, in this order: model-specific guidance first (where Fable 5, Sonnet 5, and Opus 4.8 behave differently and what to change), general techniques for all current models second, and migration considerations for prompts moving from earlier generations last. |
| Specify the exact output format tec_increase_consistency_format |
guardrail | all |
Precisely define the desired output format (JSON, XML, or a custom template) so Claude understands every formatting element required, rather than describing the shape loosely. |
| Prefill is not supported on Fable 5 — use structured outputs or system instructions instead tec_increase_consistency_no_prefill_on_fable |
guardrail | claude-fable-5 |
Prefilling the Assistant turn is a real technique for format consistency, but it is explicitly NOT supported on Claude Fable 5, Claude Mythos 5, Claude Mythos Preview, Claude Opus 4.8/4.7/4.6, or Claude Sonnet 4.6. |
| Explicitly allow "I don't know" tec_reduce_hallucinations_idk |
guardrail | all |
Give Claude explicit permission to admit uncertainty — this simple technique can drastically reduce false information. |
| Extract direct quotes before analysis on long documents tec_reduce_hallucinations_quotes |
guardrail | all |
For documents over roughly 20k tokens, ask Claude to extract word-for-word quotes first before performing its task. |
| Automatic caching needs one field; explicit breakpoints give fine control tec_prompt_caching_automatic |
infrastructure | all |
Prompt caching resumes from specific prefixes to cut processing time and cost on repeated or consistent prompts. |
| Define exactly when Fable 5 should pause vs. proceed tec_fable5_autonomy_checkpoint |
model specific | claude-fable-5 |
"Pause for the user only when the work genuinely requires them: a destructive or irreversible action, a real scope change, or input only they can provide. |
| Consider all effort levels, not just the highest tec_fable5_effort_levels |
model specific | claude-fable-5 |
Effort is the primary lever for the intelligence/latency/cost trade-off on Fable 5. |
| Lead with the outcome, not the reasoning trail tec_fable5_lead_with_outcome |
model specific | claude-fable-5 |
"Lead with the outcome. |
| Effort defaults to high; instructions are read literally, especially at low effort tec_sonnet5_effort_literalism |
model specific | claude-sonnet-5 |
On Claude Sonnet 5, effort defaults to high (same as Sonnet 4.6); raise to xhigh for the hardest coding/agentic tasks. |
| Calibrate verbosity explicitly rather than assuming a fixed style tec_sonnet5_verbosity |
model specific | claude-sonnet-5 |
Claude Sonnet 5 calibrates response length to task complexity rather than defaulting to a fixed verbosity — shorter answers on simple lookups, longer ones on open-ended analysis. |
| Accepted advisor pairings, verbatim tec_advisor_accepted_pairings |
tooling | all |
The advisor must be at least as capable as the main model. |
| Pair a Sonnet main model with a Fable 5 advisor tec_advisor_pairing_sonnet_fable |
tooling | sonnet+fable |
The advisor tool lets Claude consult a second, typically stronger model at key moments during a task — before committing to an approach, when stuck on a recurring error, or before declaring a task complete. |
| Use the Console prompt improver for complex, high-accuracy tasks tec_prompting_tools_improver |
tooling | all |
The prompt improver enhances a prompt template in four steps: locate and extract examples, produce a structured initial draft with XML tags, add chain-of-thought refinement, then enhance the examples to demonstrate the new reasoning. |
Machine-readable version: GET /api/prompt-techniques