File size: 5,116 Bytes
921d377 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 | """
Intent parsing — prompt → structured plan.
Phase 1 implementation: deterministic heuristic. Scans the prompt
for length hints, branch-count hints, topic keywords, then stitches
a full Intent from the matching PlanningPreset.
Phase 2: swap in ``backend/app/openai_compat_endpoint.run_turn``
(or any LLM) for topic extraction — the ``parse_prompt`` signature
stays stable so downstream modules don't change.
"""
from __future__ import annotations
import re
from dataclasses import dataclass, field
from typing import Any, Dict, List, Optional
from ..config import InteractiveConfig
from .audience import Audience, resolve_audience
from .presets import PlanningPreset, get_preset
@dataclass(frozen=True)
class Intent:
"""Fully-resolved planner output.
Everything downstream (branching/builder, script/generator)
works purely from this struct — no re-parsing of the prompt.
"""
prompt: str
mode: str
objective: str
audience: Audience
topic: str = ""
branch_count: int = 3
depth: int = 3
scenes_per_branch: int = 3
success_metric: str = ""
seed_intents: List[str] = field(default_factory=list)
scheme: str = "xp_level"
raw_hints: Dict[str, Any] = field(default_factory=dict)
# ─────────────────────────────────────────────────────────────────
# Heuristic hint extractors
# ─────────────────────────────────────────────────────────────────
_BRANCH_HINT = re.compile(r"(\d+)\s*(?:branches|paths|choices)", re.I)
_DEPTH_HINT = re.compile(r"(\d+)\s*(?:steps|scenes|levels)\s*deep", re.I)
_SCENES_HINT = re.compile(r"(\d+)\s*(?:scenes|clips)(?!\s*deep)", re.I)
_TOPIC_HINT = re.compile(r"(?:about|teach|explain|demonstrate)\s+([a-zA-Z][\w\s\-]{1,60}?)(?:[,\.]|$)", re.I)
_METRIC_HINT = re.compile(r"(?:until|so that|so)\s+(?:the|they|user)\s+(.{5,60}?)(?:[,\.]|$)", re.I)
def _extract_int(prompt: str, pattern: re.Pattern[str], default: int, max_: int) -> int:
m = pattern.search(prompt)
if not m:
return default
try:
v = int(m.group(1))
return max(1, min(v, max_))
except (TypeError, ValueError):
return default
def _extract_topic(prompt: str) -> str:
m = _TOPIC_HINT.search(prompt)
if not m:
return ""
return m.group(1).strip().strip('"\'')
def _extract_metric(prompt: str) -> str:
m = _METRIC_HINT.search(prompt)
if not m:
return ""
return m.group(1).strip().rstrip('.,')
# ─────────────────────────────────────────────────────────────────
# Public API
# ─────────────────────────────────────────────────────────────────
def parse_prompt(
prompt: str,
*,
cfg: InteractiveConfig,
mode: str = "sfw_general",
audience_hints: Optional[Dict[str, Any]] = None,
) -> Intent:
"""Turn a free-text prompt into a fully-resolved Intent.
Raises
------
ValueError
If ``mode`` has no matching preset AND no sane default.
"""
preset: Optional[PlanningPreset] = get_preset(mode) or get_preset("sfw_general")
if preset is None:
raise ValueError(f"No planning preset for mode '{mode}'")
audience = resolve_audience(prompt, explicit=audience_hints, default_language="en")
branch_count = _extract_int(prompt, _BRANCH_HINT, preset.default_branch_count, cfg.max_branches)
depth = _extract_int(prompt, _DEPTH_HINT, preset.default_depth, cfg.max_depth)
scenes_per_branch = _extract_int(prompt, _SCENES_HINT, preset.default_scenes_per_branch, 20)
topic = _extract_topic(prompt)
metric = _extract_metric(prompt)
# Cap total nodes at the configured ceiling — branches * depth *
# scenes_per_branch is an UPPER bound before merge-point
# collapsing, so scale branch_count down if we're over.
est_nodes = branch_count * depth * max(1, scenes_per_branch)
while est_nodes > cfg.max_nodes_per_experience and branch_count > 1:
branch_count -= 1
est_nodes = branch_count * depth * max(1, scenes_per_branch)
objective = preset.objective_template
if topic:
objective = objective.replace("{topic}", topic)
objective = objective.replace("{level}", audience.level)
return Intent(
prompt=prompt,
mode=mode,
objective=objective,
audience=audience,
topic=topic,
branch_count=branch_count,
depth=depth,
scenes_per_branch=scenes_per_branch,
success_metric=metric,
seed_intents=list(preset.seed_intents),
scheme=preset.default_scheme,
raw_hints={"preset": preset.mode, "topology": preset.default_topology},
)
|