{
  "prompt_techniques": [
    {
      "id": "tec_best_practices_structure",
      "category": "general",
      "applies_to": "all",
      "title": "The reference is organized model-specific, then general, then migration",
      "guidance": "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. This site mirrors that same ordering in its own category column: model_specific rows first, general/guardrail/tooling/infrastructure after.",
      "source_doc": "platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_best_practices_structure"
    },
    {
      "id": "tec_increase_consistency_format",
      "category": "guardrail",
      "applies_to": "all",
      "title": "Specify the exact output format",
      "guidance": "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. For guaranteed JSON schema conformance specifically, use Structured Outputs instead of prompt-only formatting — it provides guaranteed schema compliance.",
      "source_doc": "platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/increase-consistency.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_increase_consistency_format"
    },
    {
      "id": "tec_increase_consistency_no_prefill_on_fable",
      "category": "guardrail",
      "applies_to": "claude-fable-5",
      "title": "Prefill is not supported on Fable 5 — use structured outputs or system instructions instead",
      "guidance": "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. On those models, use Structured Outputs where supported, or system-prompt instructions, instead of relying on a prefill trick.",
      "source_doc": "platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/increase-consistency.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_increase_consistency_no_prefill_on_fable"
    },
    {
      "id": "tec_reduce_hallucinations_idk",
      "category": "guardrail",
      "applies_to": "all",
      "title": "Explicitly allow \"I don't know\"",
      "guidance": "Give Claude explicit permission to admit uncertainty — this simple technique can drastically reduce false information. Example instruction: \"If you're unsure about any aspect or if the report lacks necessary information, say I don't have enough information to confidently assess this.\"",
      "source_doc": "platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_reduce_hallucinations_idk"
    },
    {
      "id": "tec_reduce_hallucinations_quotes",
      "category": "guardrail",
      "applies_to": "all",
      "title": "Extract direct quotes before analysis on long documents",
      "guidance": "For documents over roughly 20k tokens, ask Claude to extract word-for-word quotes first before performing its task. This grounds the response in the actual text rather than a paraphrase, reducing hallucination. Pair with citation verification: have Claude cite a supporting quote for each claim, and retract any claim it cannot support with one.",
      "source_doc": "platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_reduce_hallucinations_quotes"
    },
    {
      "id": "tec_prompt_caching_automatic",
      "category": "infrastructure",
      "applies_to": "all",
      "title": "Automatic caching needs one field; explicit breakpoints give fine control",
      "guidance": "Prompt caching resumes from specific prefixes to cut processing time and cost on repeated or consistent prompts. Automatic caching: add a single cache_control field at the top level of the request — the system applies the cache breakpoint to the last cacheable block and moves it forward as the conversation grows. Best for multi-turn conversations. Explicit cache breakpoints: place cache_control directly on individual content blocks for fine-grained control over exactly what gets cached. Eligible for Zero Data Retention (ZDR) — under a ZDR arrangement, cached data is not stored after the response returns.",
      "source_doc": "platform.claude.com/docs/en/build-with-claude/prompt-caching.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_prompt_caching_automatic"
    },
    {
      "id": "tec_fable5_autonomy_checkpoint",
      "category": "model_specific",
      "applies_to": "claude-fable-5",
      "title": "Define exactly when Fable 5 should pause vs. proceed",
      "guidance": "\"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. If you hit one of these, ask and end the turn, rather than ending on a promise.\" For autonomous pipelines specifically: \"You are operating autonomously... For reversible actions that follow from the original request, proceed without asking.\"",
      "source_doc": "platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_fable5_autonomy_checkpoint"
    },
    {
      "id": "tec_fable5_effort_levels",
      "category": "model_specific",
      "applies_to": "claude-fable-5",
      "title": "Consider all effort levels, not just the highest",
      "guidance": "Effort is the primary lever for the intelligence/latency/cost trade-off on Fable 5. Use high as the default, xhigh for the most capability-sensitive workloads, and medium or low for routine work. Lower effort levels on Fable 5 still perform well and often exceed xhigh performance on prior models — on routine work at higher effort, Fable 5 can gather context and deliberate beyond what the task needs.",
      "source_doc": "platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_fable5_effort_levels"
    },
    {
      "id": "tec_fable5_lead_with_outcome",
      "category": "model_specific",
      "applies_to": "claude-fable-5",
      "title": "Lead with the outcome, not the reasoning trail",
      "guidance": "\"Lead with the outcome. Your first sentence after finishing should answer what happened or what did you find. Supporting detail and reasoning come after. Being readable and being concise are different things, and readability matters more. The way to keep output short is to be selective about what you include (drop details that don't change what the reader would do next), not to compress the writing into fragments, abbreviations, arrow chains, or jargon.\" A short brevity instruction like this is as effective as enumerating every unwanted pattern by name.",
      "source_doc": "platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_fable5_lead_with_outcome"
    },
    {
      "id": "tec_sonnet5_effort_literalism",
      "category": "model_specific",
      "applies_to": "claude-sonnet-5",
      "title": "Effort defaults to high; instructions are read literally, especially at low effort",
      "guidance": "On Claude Sonnet 5, effort defaults to high (same as Sonnet 4.6); raise to xhigh for the hardest coding/agentic tasks. Sonnet 5 interprets prompts literally and explicitly — it does not silently generalize an instruction from one item to another. If an instruction should apply broadly, say so explicitly (e.g. \"apply this formatting to every section, not just the first one\"). Manual extended thinking (budget_tokens) is removed on Sonnet 5 and returns a 400 error; use adaptive thinking with the effort parameter instead.",
      "source_doc": "platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-sonnet-5.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_sonnet5_effort_literalism"
    },
    {
      "id": "tec_sonnet5_verbosity",
      "category": "model_specific",
      "applies_to": "claude-sonnet-5",
      "title": "Calibrate verbosity explicitly rather than assuming a fixed style",
      "guidance": "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. If a product depends on a specific style, add an explicit instruction, e.g. \"Provide concise, focused responses. Skip non-essential context, and keep examples minimal.\" Positive examples of the desired style outperform negative instructions (\"don't do X\").",
      "source_doc": "platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-sonnet-5.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_sonnet5_verbosity"
    },
    {
      "id": "tec_advisor_accepted_pairings",
      "category": "tooling",
      "applies_to": "all",
      "title": "Accepted advisor pairings, verbatim",
      "guidance": "The advisor must be at least as capable as the main model. Accepted pairings: Haiku 4.5 main -> Fable, Opus, or Sonnet advisor (Haiku can call the advisor but cannot act as one). Sonnet 4.6 main -> Fable, Opus, or Sonnet advisor. Opus 4.6+ main -> Fable or Opus at or above the main model's version (an Opus 4.7 main with an Opus 4.6 advisor is rejected). Fable 5 main (v2.1.170+) -> Fable advisor only (an Opus or Sonnet advisor is rejected, since Fable is a higher tier than both).",
      "source_doc": "code.claude.com/docs/en/advisor.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_advisor_accepted_pairings"
    },
    {
      "id": "tec_advisor_pairing_sonnet_fable",
      "category": "tooling",
      "applies_to": "sonnet+fable",
      "title": "Pair a Sonnet main model with a Fable 5 advisor",
      "guidance": "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. \"Sonnet main + Fable advisor: Fable 5 guidance at decision points without running Fable 5 throughout. Requires v2.1.170 or later and Fable 5 access.\" This is the pairing this repo dogfoods: this session runs on Claude Sonnet 5 as the main model, and the actual subagentcode.com and subagentprompts.com builds each used a real Agent-tool dispatch to Fable 5 — once as the primary builder, once as an advisor reviewing the draft — rather than running Fable 5 for the entire session.",
      "source_doc": "code.claude.com/docs/en/advisor.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_advisor_pairing_sonnet_fable"
    },
    {
      "id": "tec_prompting_tools_improver",
      "category": "tooling",
      "applies_to": "all",
      "title": "Use the Console prompt improver for complex, high-accuracy tasks",
      "guidance": "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. Best for complex tasks requiring detailed reasoning and where accuracy matters more than speed or latency; for latency- or cost-sensitive applications, prefer simpler prompts instead.",
      "source_doc": "platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-tools.md",
      "source_site": "subagentprompts.com",
      "created_at": "2026-07-01 20:54:11",
      "cite_as": "https://subagentprompts.com/api/prompt-techniques/tec_prompting_tools_improver"
    }
  ]
}