intent-parser
Installation
SKILL.md
intent-parser
Maps a Telegram message to one of six intents, plus extracted parameters. Lightweight regex layer + Claude classification fallback. Output is a JSON payload that downstream skills consume.
The six intents
| Intent | Trigger shape | Fields extracted | Routes to |
|---|---|---|---|
research |
Question form, "what / find / research / look up / check the latest" | topic, time_budget (quick/deep), use_corpus |
Margot via margot-bridge |
ticket |
"file / create / open / track" + project/title hint | team, project, title, priority |
Dispatcher → Linear MCP |
reply |
"tell / message / reply / draft / respond to" + recipient | recipient, medium (telegram/email/linear), body_hint |
Scribe → telegram-draft-for-review |
reminder |
"remind / nudge / chase me / on Thursday" + time | when (ISO-8601 absolute), what |
Dispatcher (Wave 2 calendar) |
flow |
Multi-step ("first do X then Y"), or explicit "run the flow" | steps[] (3+ ordered actions) |
Dispatcher (dispatcher-core) |
unknown |
Doesn't match above | raw message | CoS clarification (one Telegram follow-up) |