AgentSkillsCN

thesis-classify

为单个信号运行关键词+LLM论文分类，并将其存储于thesis_classifications中。

SKILL.md
--- frontmatter
name: thesis-classify
description: Run keyword + LLM thesis classification for a single signal and store
  it in thesis_classifications.
user-invocable: true
allowed-tools:
- Bash
- Read

thesis-classify

When to use

  • •You want a high-signal LLM-based category/thesis fit estimate for an example signal.
  • •You’re debugging keyword vs LLM disagreements.

Inputs

  • •signal_id
  • •LLM model (default gemini-2.0-flash)
  • •GEMINI_API_KEY env var (google-genai)

Workflow

  1. •Run: python -m ops.cli quality thesis-classify <signal_id> --model gemini-2.0-flash.
  2. •Confirm a row exists in thesis_classifications for that signal (latest classification).

Outputs

  • •JSON summary of classification fields (keyword_score, thesis_match, thesis_fit_score, category, etc).

Guardrails

  • •LLM output should not be treated as ground truth; use it as a decision aid + for tuning.
  • •Handle API failures gracefully; do not spam retries.

References

  • •references/reference.md
  • •docs/QUALITY_OPS_ARCHITECTURE.md