AGENT SKILL

systemone

Use when the user asks to score, rate, triage, or prioritise a situation with 9router systemone. Returns confidence 0-1 for predefined questions — not chat text.

License 许可
MIT
最近更新
2026-11-06
Stars
★ 0
Forks
0

01 前言 Front Matter

name
systemone
description
Use when the user asks to score, rate, triage, or prioritise a situation with 9router systemone. Returns confidence 0-1 for predefined questions — not chat text.

02 资源文件 1

03 SKILL.md 正文

systemone — structured scoring via 9router

Scores a situation against predefined questions and returns a confidence 0-1 per question. It does NOT return prose, does not stream, and has no tools.

When this is the wrong tool

If the user wants a reply written, code edited, files read, or a command run, this endpoint cannot do it — use the normal chat model. Reach for systemone only when the goal is a number (triage, urgency, confidence, priority).

Usage

python3 ~/.paseo/skills/systemone.py "<situation text>" \
  -q urgent="Does this request require urgent attention?" \
  -q angry="Is the customer angry?"

Output:

model: jev-1.13-free
  urgent         0.950
  angry          0.930
usage: 304 in / 37 out   cost: 0

Use -m <model> to switch models. Env overrides: SYSTEMONE_URL, SYSTEMONE_KEY, SYSTEMONE_MODEL.

Endpoint & Types

Endpoint: http://127.0.0.1:8008/v1/systemone (Model: kev-latest).

Payload structure requires state (context string) and questions (map of question objects).

Supported question types:

  1. noul: Binary Yes/No confidence 0.0 - 1.0.
    • Schema: {"type": "noul", "instructions": "..."}
    • Response: {"type": "noul", "noul": <float>}
  2. choice: Multi-class routing/classification.
    • Schema: {"type": "choice", "instructions": "...", "criteria": {"<key>": "<desc>"}}
    • Response: {"type": "choice", "choice": "<key>", "confidence": <float>, "probabilities": {...}}
  3. score: Discrete/continuous milestone evaluation.
    • Schema: {"type": "score", "instructions": "...", "criteria": ["0: ...", "1: ..."]}
    • Response: {"type": "score", "score": <float>, "legend": {...}}

Constraints

  • stream, tools, system, and messages rejected with 400.
  • Stateless: pass complete context in state.

Reading the score

Scores are model confidence, not truth. 0.9 means the model is confident, not that the situation is 90% urgent. Two independent questions scored together are cheaper and more stable than one repeated call — a single call answers many questions, and repeat-call variance is high (same input has been observed at 0.43 and 0.95 on different questions).

04 同仓库更多 Skills