Download src/polyhedra/core/mention_parser.py from polyHedra/PolyHedra: direct link, hf CLI and curl.
- Browser
- Download file 2.48 kB
-
https://huggingface.co/spaces/polyHedra/PolyHedra/resolve/main/src/polyhedra/core/mention_parser.py
- Command line
-
hf download hf://spaces/polyHedra/PolyHedra/src/polyhedra/core/mention_parser.py
-
curl -L -o mention_parser.py https://huggingface.co/spaces/polyHedra/PolyHedra/resolve/main/src/polyhedra/core/mention_parser.py
2.48 kB
| """@mention parser for Polyhedra v3.0. | |
| Detects ``@<role>`` patterns in incoming messages and resolves them to | |
| known roles from the active session's RoleRegistry. | |
| Supports both plain-text (@scout) and Slack-style (<@USERID>) formats. | |
| The Slack-specific resolution layer maps user/bot IDs to role names and is | |
| injected by the messaging integration layer. | |
| """ | |
| from __future__ import annotations | |
| import re | |
| from dataclasses import dataclass | |
| from polyhedra.agents.base import Mention | |
| from polyhedra.core.role_registry import RoleRegistry | |
| # Matches @word or @word-word (hyphenated) | |
| _AT_MENTION_RE = re.compile(r"@([\w-]+)", re.IGNORECASE) | |
| class ParsedMention: | |
| """The result of parsing a raw message for @mentions.""" | |
| mentioned_roles: list[str] # resolved role names (lowercase) | |
| human_addressed: bool # True if message is directed at the human (held role) | |
| raw_text: str | |
| class MentionParser: | |
| """ | |
| Extract @mention role references from raw message text. | |
| The parser is platform-agnostic: it just looks for ``@<word>`` patterns. | |
| Platform adapters (Slack, Teams) must normalise their mention syntax | |
| (e.g. ``<@U12345>``) before passing to ``parse()``. | |
| """ | |
| def __init__(self, registry: RoleRegistry) -> None: | |
| self._registry = registry | |
| def parse(self, text: str, from_role: str = "human") -> list[Mention]: | |
| """ | |
| Parse all @mentions from ``text`` and return a list of Mention objects. | |
| Unrecognised role names are silently ignored. | |
| """ | |
| matches = _AT_MENTION_RE.findall(text) | |
| mentions: list[Mention] = [] | |
| seen: set[str] = set() | |
| for raw_role in matches: | |
| role = raw_role.lower() | |
| if role in seen: | |
| continue | |
| seen.add(role) | |
| if self._registry.get(role) is not None: | |
| mentions.append( | |
| Mention( | |
| mentioned_role=role, | |
| from_role=from_role, | |
| content=text, | |
| ) | |
| ) | |
| return mentions | |
| def extract_roles(self, text: str) -> list[str]: | |
| """Return just the list of recognised role names mentioned in text.""" | |
| return [m.mentioned_role for m in self.parse(text)] | |
| def strip_mentions(self, text: str) -> str: | |
| """Remove @mention tokens from text (useful before feeding to LLM).""" | |
| return _AT_MENTION_RE.sub("", text).strip() | |