petal/internal/llm/prompts.go

package llm

import "fmt"

// checkpointSystemPrompt is the grammar-checkpoint instruction. It asks for
// strict JSON (no fences, no preamble) so Complete's output parses directly.
const checkpointSystemPrompt = `You are a warm, encouraging writing assistant helping someone who speaks English as a second language. ` +
	`Analyze the text below and identify up to 5 issues: grammar errors, unnatural phrasing, ` +
	`incorrect idiom usage, or unclear sentences that are common ESL patterns.

Be specific, friendly, and explain WHY each suggestion improves the writing.%s

Respond ONLY with valid JSON. No preamble, no markdown fences. Format:
{
  "suggestions": [
    {
      "original": "exact text from the document that needs fixing",
      "replacement": "corrected version",
      "explanation": "friendly one-sentence explanation",
      "type": "grammar|phrasing|idiom|clarity"
    }
  ]
}

If the writing looks good, return: {"suggestions": []}`

// toneGuidance returns a sentence steering the checkpoint toward the writer's
// chosen tone, or "" for the neutral default. The clause is appended to the
// checkpoint instructions so the model's phrasing suggestions fit the target
// register (e.g. an academic essay vs a casual journal). Unknown values fall
// back to no steering, so a stray tone string is harmless.
func toneGuidance(tone string) string {
	clause, ok := map[string]string{
		"academic":     "formal, academic, and objective — suited to a school essay or research paper",
		"professional": "polished and professional — suited to a workplace email or report",
		"casual":       "relaxed, friendly, and conversational",
		"humorous":     "light, playful, and good-humored",
		"creative":     "vivid, expressive, and imaginative — suited to a story or personal narrative",
		"persuasive":   "confident and persuasive — suited to an argument or opinion piece",
	}[tone]
	if !ok {
		return ""
	}
	return "\n\nThe writer wants this document to read as " + clause + ". When phrasing could " +
		"be improved, prefer suggestions that fit that tone, and gently flag wording that clashes with it."
}

// pairCheckpointSystemPrompt is the grammar checkpoint for a document written in
// the writer's own language rather than in English.
//
// It is a separate constant rather than a language clause appended to
// checkpointSystemPrompt, because that prompt opens by naming the reader as an
// ESL learner and asks for "common ESL patterns" — appending "and explain in
// Portuguese" would hand the model two contradictory framings. Only the framing
// differs; the JSON contract and the tone clause below it are the same
// instructions in the same order, so the two prompts stay comparable.
//
// The "never translate" line is the one the model most wants to disobey: asked
// to improve Portuguese while being an English writing assistant by training, it
// will happily hand back an English rendering, which is a translation card
// (Phase 25's `isTranslation`) and not a correction.
const pairCheckpointSystemPrompt = `You are a warm, encouraging writing assistant. The person you are helping is ` +
	`writing in %[1]s, and the text below is %[1]s. ` +
	`Analyze it and identify up to 5 issues: grammar errors, unnatural phrasing, ` +
	`incorrect idiom usage, or unclear sentences.

Both "original" and "replacement" must be written in %[1]s. You are improving their %[1]s writing — ` +
	`never translate it into English, and never suggest they write in English instead.
Write every "explanation" in %[2]s.

Be specific, friendly, and explain WHY each suggestion improves the writing.%[3]s

Respond ONLY with valid JSON. No preamble, no markdown fences. Format:
{
  "suggestions": [
    {
      "original": "exact text from the document that needs fixing",
      "replacement": "corrected version",
      "explanation": "friendly one-sentence explanation",
      "type": "grammar|phrasing|idiom|clarity"
    }
  ]
}

If the writing looks good, return: {"suggestions": []}`

// CheckpointMessages builds the message array for a grammar checkpoint over the
// given (already-truncated) document text, steered toward the document's tone
// and aimed at the language the document is actually written in.
func CheckpointMessages(contentText, tone string, t Target) []Message {
	system := fmt.Sprintf(checkpointSystemPrompt, toneGuidance(tone))
	if t.Flipped() {
		system = fmt.Sprintf(pairCheckpointSystemPrompt, t.Correct.Name, t.Explain.Name, toneGuidance(tone))
	}
	return []Message{
		{Role: "system", Content: system},
		{Role: "user", Content: contentText},
	}
}

// voiceSystemPrompt drives the Tier-1 voice-consistency pass. It is a distinct
// pass from the grammar checkpoint (spec: "do not bundle them") — the model
// reads the whole document to learn the writer's natural voice, then flags
// passages that read as tonally out of place. `replacement` is null: these are
// awareness-only, with no correction to apply.
const voiceSystemPrompt = `You are a warm, encouraging writing assistant helping someone who speaks English as a second language. ` +
	`You are reviewing a COMPLETE document for VOICE CONSISTENCY only — not grammar.

Read the whole document to learn the writer's natural voice, then identify any passages (2 or more sentences) ` +
	`that feel tonally inconsistent with the surrounding writing — unusually formal, unusually polished, or phrased ` +
	`in a way that differs from the writer's established voice elsewhere in the document. These often signal text ` +
	`that was paraphrased too closely from another source. Do not flag the first paragraph (there is no baseline yet). ` +
	`Do not flag grammar or spelling mistakes — only voice.

Respond ONLY with valid JSON. No preamble, no markdown fences. Format:
{
  "suggestions": [
    {
      "original": "exact passage from the document that feels inconsistent",
      "replacement": null,
      "explanation": "friendly one-sentence note, e.g. 'This passage sounds more formal than the rest of your writing — worth reviewing.'",
      "type": "voice"
    }
  ]
}

If the voice is consistent throughout, return: {"suggestions": []}`

// pairVoiceSystemPrompt is the voice pass for a document in the writer's own
// language. Voice consistency is the one pass that transfers across languages
// unchanged — a paragraph that reads as pasted from elsewhere reads that way in
// any language — so only the framing and the explanation language move.
const pairVoiceSystemPrompt = `You are a warm, encouraging writing assistant. The person you are helping is writing ` +
	`in %[1]s. You are reviewing a COMPLETE %[1]s document for VOICE CONSISTENCY only — not grammar.

Read the whole document to learn the writer's natural voice, then identify any passages (2 or more sentences) ` +
	`that feel tonally inconsistent with the surrounding writing — unusually formal, unusually polished, or phrased ` +
	`in a way that differs from the writer's established voice elsewhere in the document. These often signal text ` +
	`that was paraphrased too closely from another source. Do not flag the first paragraph (there is no baseline yet). ` +
	`Do not flag grammar or spelling mistakes — only voice.

Quote each passage exactly as it appears, in %[1]s. Write every "explanation" in %[2]s.

Respond ONLY with valid JSON. No preamble, no markdown fences. Format:
{
  "suggestions": [
    {
      "original": "exact passage from the document that feels inconsistent",
      "replacement": null,
      "explanation": "friendly one-sentence note about why this passage sounds unlike the rest",
      "type": "voice"
    }
  ]
}

If the voice is consistent throughout, return: {"suggestions": []}`

// VoiceMessages builds the message array for a voice-consistency pass. Unlike
// the checkpoint, the caller passes the WHOLE document (no truncation) — voice
// consistency is judged against the established voice everywhere else.
//
// The pass had no language argument at all before Phase 28, which was the same
// English assumption the checkpoint made, just unstated.
func VoiceMessages(contentText string, t Target) []Message {
	system := voiceSystemPrompt
	if t.Flipped() {
		system = fmt.Sprintf(pairVoiceSystemPrompt, t.Correct.Name, t.Explain.Name)
	}
	return []Message{
		{Role: "system", Content: system},
		{Role: "user", Content: contentText},
	}
}

// collocationSystemPrompt drives the collocation coach — the single most
// valuable polish for an ESL writer. It flags word PAIRINGS that are not wrong,
// just non-native ("do a decision" → "make a decision", "strong rain" → "heavy
// rain"), and explicitly DEFERS real grammar/spelling errors to the grammar
// checkpoint so the two families don't overlap. Every explanation is framed as a
// warm "natives usually say…" note with a short gloss in the writer's own
// language — never "error/wrong" — because these are stylistic, not mistakes. It is a distinct
// pass from the grammar checkpoint (do not bundle them). `replacement` carries
// the natural pairing the writer can accept in one tap.
const collocationSystemPrompt = `You are a warm, encouraging writing assistant helping someone who speaks English as a second language. ` +
	`You are reviewing a COMPLETE document for COLLOCATIONS only — the natural word pairings native speakers use.

A collocation is a pair or short group of words that native speakers habitually say together. ESL writers ` +
	`often choose words that are grammatically correct but sound non-native: "do a decision" instead of "make a decision", ` +
	`"strong rain" instead of "heavy rain", "say a joke" instead of "tell a joke". These are NOT grammar mistakes — they ` +
	`are just not what a native speaker would naturally say.

Identify up to 5 such non-native word pairings. For each, give the natural pairing a native speaker would use. ` +
	`Be gentle and specific. Do NOT flag grammar errors, spelling mistakes, or unclear sentences — those are handled ` +
	`elsewhere. Only flag word pairings that are correct but sound non-native.%s

Phrase every explanation warmly as "Natives usually say…" and include a brief %s gloss in parentheses. ` +
	`Never use the words "error", "wrong", or "mistake" — these are friendly polish, not corrections.

Respond ONLY with valid JSON. No preamble, no markdown fences. Format:
{
  "suggestions": [
    {
      "original": "exact word pairing from the document",
      "replacement": "the natural native pairing",
      "explanation": "friendly note, e.g. 'Natives usually say \"make a decision\" rather than \"do a decision\" (native usage / 地道说法).'",
      "type": "collocation"
    }
  ]
}

If every pairing already sounds natural, return: {"suggestions": []}`

// CollocationMessages builds the message array for a collocation pass over the
// WHOLE document (no truncation), gently steered toward the document's tone so a
// hint can prefer a register-appropriate pairing. The parenthetical gloss is
// written in the writer's own language — `Pair`, not `Explain`: the gloss is
// addressed to her rather than to the document.
//
// The coach itself remains English-only. Collocation lists are the one thing
// here that is genuinely per-language knowledge rather than framing, and
// "natives usually say" for Portuguese is a claim this prompt has no grounds to
// make yet; a flipped document simply gets the pass it always got. (Phase 28
// moved the checkpoint and the voice pass; this one waits for evidence.)
func CollocationMessages(contentText, tone string, t Target) []Message {
	return []Message{
		{Role: "system", Content: fmt.Sprintf(collocationSystemPrompt, toneGuidance(tone), t.Pair.Name)},
		{Role: "user", Content: contentText},
	}
}

// askPetalSystemTemplate is the Ask Petal tutor prompt. The suggestion context
// is interpolated in; the user's own messages are appended after this system
// turn by the caller.
//
// The reply is bilingual, the pair language first. Until UX item 6 it mirrored
// the language of the question instead — self-consistent, but it meant asking in
// one language cost you the other, and the writer doesn't always know which one
// the answer will be clearer in. Which half is the safety net and which is the
// lesson depends on who is writing: the pair is (English + X) either way, and an
// English speaker learning French wants the French half for the same reason a
// Mandarin speaker learning English wants the English one. Petal cannot tell
// them apart from a chat message, and doesn't need to — every other explanation
// surface already gives both (the card's English body, the seeded bubble in the
// pair language). The answer that goes deepest into the "why" was the one place
// that didn't.
//
// The blank line between the halves is a contract with the client: AskPetal.tsx
// splits on the first one to render her language prominently and the English
// beneath it, mirroring the companion's bubble. A model that ignores the
// instruction and writes one language degrades to a single plain block — the
// answer is still readable, which is why the split is a rendering nicety and
// never a parse the reply depends on.
const askPetalSystemTemplate = `You are Petal, a warm and patient English writing tutor helping someone who is learning English ` +
	`as a second language. You are currently discussing a specific writing suggestion.

Suggestion context:
- Original text: "%[1]s"
- Suggested replacement: "%[2]s"
- Issue type: %[3]s
- Initial explanation: "%[4]s"
- Surrounding paragraph: "%[5]s"

The user wants to understand this suggestion better. Answer in BOTH languages, every time, ` +
	`whichever language they asked their question in: first the whole answer in %[6]s, then the ` +
	`same answer again in English. Separate the two with a single blank line. Do not label them, ` +
	`do not use a blank line anywhere else, and do not mix the two languages within one half — ` +
	`each half is complete on its own.

One of those two languages is the one they are surest in and the other is the one they are ` +
	`working in — you do not know which way round, so give both and let them choose. Both halves ` +
	`say the same thing: do not put a point in one that is missing from the other.

Explain clearly and kindly. Use simple language appropriate to the user's message. Give examples ` +
	`when helpful. If they ask "why" (or "%[7]s"), explain the grammar rule or idiom behind it. ` +
	`If they suggest an alternative phrasing, evaluate it honestly.

Keep each half concise (2-3 sentences). This is a chat, not an essay. Be encouraging — ` +
	`learning a language is hard and they're doing great.`

// AskPetalSystemPrompt fills the tutor prompt with one suggestion's context and
// the writer's pair language, which is the one she may ask her question in.
func AskPetalSystemPrompt(original, replacement, suggestionType, explanation, paragraph string, lang Lang) string {
	return fmt.Sprintf(askPetalSystemTemplate, original, replacement, suggestionType, explanation, paragraph, lang.Name, lang.Why)
}

// rewriteSystemTemplate drives the "say it more naturally" / tone-rewrite tool.
// The writer selects a passage and picks a style; the model rewrites that
// passage in place. The instruction is deliberately strict about returning ONLY
// the rewritten passage so the result can be dropped straight into the editor —
// no quotes, no preamble, no commentary to strip.
const rewriteSystemTemplate = `You are Petal, a warm English writing assistant helping someone who speaks English ` +
	`as a second language. Rewrite the passage the user sends so that it %s, while preserving its original ` +
	`meaning. Fix any grammar mistakes and awkward phrasing along the way. Keep it about the same length — ` +
	`do not add new ideas, explanations, or commentary.

Respond with ONLY the rewritten passage. No quotation marks around it, no preamble, no notes — just the ` +
	`rewritten English text, ready to drop back into the document.`

// styleGuidance maps a rewrite style onto the clause describing the target
// register. "natural" is the default "say it more naturally" action; the rest
// mirror the document-tone vocabulary (see toneGuidance / the ToneSelect UI).
// An unknown style falls back to the natural rewrite.
func styleGuidance(style string) string {
	switch style {
	case "academic":
		return "reads as formal, academic English suited to a school essay or research paper"
	case "professional":
		return "reads as polished, professional English suited to a workplace email or report"
	case "casual":
		return "sounds relaxed, friendly, and conversational"
	case "humorous":
		return "has a light, playful, good-humored tone"
	case "creative":
		return "is vivid, expressive, and imaginative"
	case "persuasive":
		return "is confident and persuasive"
	default: // "natural"
		return "sounds natural and fluent, the way a native English speaker would naturally say it"
	}
}

// RewriteMessages builds the message array for a tone-rewrite: the styled system
// instruction plus the passage to rewrite as the user turn.
func RewriteMessages(text, style string) []Message {
	return []Message{
		{Role: "system", Content: fmt.Sprintf(rewriteSystemTemplate, styleGuidance(style))},
		{Role: "user", Content: text},
	}
}

// translateSystemPrompt drives the explanation translator: it renders a
// suggestion's English explanation into the writer's own language so an ESL
// reader sees the "why" in her first language. Strict about returning ONLY the
// translation (no quotes, no romanisation, no English echo) so it can drop
// straight into the chat bubble. Kept warm and plain — these are short, friendly
// one-liners.
const translateSystemPrompt = `You are Petal, a warm writing assistant. Translate the English text the user ` +
	`sends into natural, friendly %[1]s. It is a short explanation of a writing ` +
	`suggestion, written for a native %[1]s speaker learning English.

Respond with ONLY the %[1]s translation. No quotation marks, no romanisation, no English, no preamble — ` +
	`just the translated sentence.`

// TranslateMessages builds the message array for translating one short English
// explanation into the writer's own language.
func TranslateMessages(text string, lang Lang) []Message {
	return []Message{
		{Role: "system", Content: fmt.Sprintf(translateSystemPrompt, lang.Name)},
		{Role: "user", Content: text},
	}
}