Prompt Library Index

# Prompt Library Index

Auto-generated by `python cli.py index`. Do not edit by hand.

**15 prompts** across 7 categories.

| Name | Category | Version | Intent |
| --- | --- | --- | --- |
| [map-reduce-summarizer](prompts/agent-and-chain/map-reduce-summarizer.prompt.md) | `agent-and-chain` | 1.0.0 | A two-stage chain for summarising a long document that exceeds a comfortable context budget: a per-chunk MAP prompt and a REDUCE prompt that merges the chunk summaries without re-reading the source. |
| [research-plan-then-act](prompts/agent-and-chain/research-plan-then-act.prompt.md) | `agent-and-chain` | 1.0.0 | Drive a tool-using agent through an explicit plan -> act -> reflect loop so it decomposes a research task, uses tools deliberately, and knows when to stop instead of looping forever. |
| [intent-classifier](prompts/classification/intent-classifier.prompt.md) | `classification` | 1.3.0 | Classify a short user message into one of a fixed set of intents, returning the label, a calibrated confidence, and a one-line rationale, with a dedicated "unknown" class so out-of-scope inputs are not force-fit. |
| [pii-redaction-gate](prompts/classification/pii-redaction-gate.prompt.md) | `classification` | 1.0.0 | A safety gate that decides whether a text contains personal data and, if so, returns a redacted copy. Built to fail safe — when uncertain it redacts rather than leaks. |
| [answer-quality-rubric](prompts/eval-rubrics/answer-quality-rubric.prompt.md) | `eval-rubrics` | 1.0.0 | A pairwise LLM-as-judge rubric for A/B prompt comparison: given a task and two candidate answers, decide which is better on a fixed set of weighted criteria, with position-bias mitigation built in. |
| [faithfulness-judge](prompts/eval-rubrics/faithfulness-judge.prompt.md) | `eval-rubrics` | 1.0.0 | An LLM-as-judge rubric that scores whether a summary or answer is faithful to a source — i.e. contains no claims unsupported by the source — returning a numeric score plus the specific unsupported spans. |
| [contact-extraction-v1-baseline](prompts/extraction/contact-extraction-v1-baseline.prompt.md) | `extraction` | 1.0.0 | Baseline variant. Extract contact details (name, email, phone, company) from free text into JSON. Minimal instructions — used as the control in the A/B eval against the hardened v2. |
| [contact-extraction-v2-hardened](prompts/extraction/contact-extraction-v2-hardened.prompt.md) | `extraction` | 2.1.0 | Hardened variant. Extract contact details into a strict JSON object with an explicit null policy, output-only-JSON guard, and normalisation rules. This is the challenger that beats v1 in the A/B eval. |
| [plain-language-simplifier](prompts/rewriting/plain-language-simplifier.prompt.md) | `rewriting` | 1.0.0 | Rewrite dense or jargon-heavy text at a target reading level without dropping any load-bearing detail, and flag (not invent) anything that cannot be simplified safely. |
| [prompt-improver](prompts/rewriting/prompt-improver.prompt.md) | `rewriting` | 1.0.0 | Rewrite a raw, under-specified prompt (or a block of free text pasted from an inbox or notes) into a stronger, structured system prompt — adding the role, context, output spec, constraints, success criteria, and refusal policy the original was missing, without changing the user's actual intent. |
| [tone-shift-rewriter](prompts/rewriting/tone-shift-rewriter.prompt.md) | `rewriting` | 1.0.0 | Rewrite a message into a target tone (e.g. warmer, more formal, more concise) while preserving its facts, commitments, and meaning — changing the delivery, never the substance. |
| [senior-code-reviewer](prompts/role-prompts/senior-code-reviewer.prompt.md) | `role-prompts` | 1.0.0 | Prime the model as a pragmatic senior engineer doing a code review that reports only high-signal issues, ranked by severity, with a concrete fix — and explicitly stays silent when there is nothing worth saying. |
| [support-triage-agent](prompts/role-prompts/support-triage-agent.prompt.md) | `role-prompts` | 1.2.0 | Prime the model as a tier-1 support triage specialist that classifies an inbound ticket, assigns urgency, and drafts a first response — without inventing facts or making promises it cannot keep. |
| [function-call-router](prompts/structured-output/function-call-router.prompt.md) | `structured-output` | 1.1.0 | Given a user utterance and a list of available tools, emit a single JSON object naming the tool to call and its arguments — or explicitly decline when no tool fits, instead of forcing a bad call. |
| [invoice-to-json](prompts/structured-output/invoice-to-json.prompt.md) | `structured-output` | 2.0.0 | Convert a free-text invoice or receipt into a strict JSON object matching a fixed schema, emitting null for any field that is genuinely absent rather than guessing. |