RSS 资讯菜单

Use this skill to turn configured RSS feeds into a preference-aware, mobile-friendly Chinese news menu. Treat the filesystem as the working memory: configuration, preferences, run history, cached raw results, output templates, and known pitfalls all live beside the skill or under the user's cache directory.

Resolve Skill Directory

Before reading or writing skill files, resolve the current skill directory. Prefer explicit environment variables, then common personal and project-level locations:

SKILL_DIR="${RSS_READER_SKILL_DIR:-}"
if [ -z "$SKILL_DIR" ]; then
  for dir in \
    "$PWD" \
    "$PWD/.claude/skills/rss-reader" \
    "$PWD/.codex/skills/rss-reader" \
    "${CLAUDE_HOME:-$HOME/.claude}/skills/rss-reader" \
    "${CODEX_HOME:-$HOME/.codex}/skills/rss-reader" \
    "${HERMES_HOME:-$HOME/.hermes}/skills/rss-reader" \
    "${OPENCLAW_HOME:-$HOME/.openclaw}/skills/rss-reader"; do
    if [ -f "$dir/SKILL.md" ] && [ -f "$dir/scripts/rss_reader.py" ]; then
      SKILL_DIR="$dir"
      break
    fi
  done
fi

If no directory is found, ask the user where the skill is installed or ask them to set RSS_READER_SKILL_DIR.

Workflow

Check preferences in preferences.yaml. If missing, empty, or initialized is not true, ask the three short setup questions from preferences.example.yaml and write the answers locally.
Run python3 "$SKILL_DIR/scripts/rss_reader.py" to fetch RSS items. The script reads references/config.yaml, deduplicates recently seen links, and writes the latest raw result to the cache.
Summarize the raw result according to preferences.yaml. Use references/output-format.md as the default first-round menu and expansion format.
When the user asks to expand, compare, view a demoted category, change preferences, or use this result in another workflow, load only the relevant cached raw result and reference file.
After a recurring failure or useful correction, update references/pitfalls.md with the symptom, cause, and preferred handling.

Filesystem Context

references/config.yaml: RSS source list and grouping config.
preferences.yaml: local personal preferences. Do not commit real user preferences.
preferences.example.yaml: preference template and first-run defaults.
references/output-format.md: default output shapes for menu, expansion, comparison, and demoted categories.
references/workflow-contract.md: input/output contract for orchestration by other agents or workflows.
references/pitfalls.md: living library of known mistakes and recovery patterns.
scripts/rss_reader.py: deterministic fetch, parse, dedupe, and cache script.
~/.cache/rss-reader/rss-history.json: dedupe history.
~/.cache/rss-reader/latest.json: latest normalized raw result for follow-up expansion.

Preference Setup

Ask only these three questions on first use:

第一次使用，我先记一下你的资讯偏好。之后会按这个筛选。

1. 你最关心哪几类？
可选：AI产品工具 / Agent开发 / 模型论文 / 公司商业 / 监管社会 / 开源项目 / AI安全 / 行业应用

2. 你想少看或不看哪几类？
可选：论文 / 融资 / 社媒热帖 / 海外长文 / 开发工具 / 监管 / 纯产品发布

3. 推送希望多短？
可选：极短 / 标准 / 稍详细

Write the result as YAML:

initialized: true
focus:
  - Agent开发
  - AI产品工具
avoid:
  - 社媒热帖
must_include: []
summary_style: detailed
max_focus_topics: 5
max_other_titles: 10

Operating Guidance

Provide useful defaults instead of rigidly enforcing a format. If the user asks for a different shape, adapt while keeping the same RSS evidence base.
Teach only the workflow-specific parts: where files are, how data is fetched, how preferences are applied, and how follow-up expansion works.
Keep first-round output as a menu unless the user explicitly asks for a report, deep analysis, links, or a workflow handoff.
Use Chinese by default for user-facing output.
Do not copy long RSS text verbatim. Summarize in your own words and include links only when expanding or when the user asks.
If the feed result is empty, explain whether it is because of dedupe, fetch failures, or genuinely no parsed items. Use references/pitfalls.md for recovery.