topic-monitor（主题监视器）

Topic Monitor

Monitor topics continuously and alert only when something looks relevant.

What’s new in v1.5.0

• RSS/Atom feed monitoring as a first-class source via feeds
• GitHub release monitoring via github_repos → https://github.com/{owner}/{repo}/releases.atom
• Feed auto-discovery from normal URLs
• OPML import for feed lists
• Advanced filters with required_keywords and exclude_keywords
• Sentiment analysis on findings: positive, negative, neutral, mixed
• Sentiment shift alerts with alert_on_sentiment_shift

Quick start

python3 scripts/quick.py "AI Model Releases"
python3 scripts/quick.py "OpenClaw Releases" --github-repos "openclaw/openclaw"
python3 scripts/quick.py "Security Advisories" --feeds "https://example.com/security.xml"

Then test it:

python3 scripts/monitor.py --dry-run --verbose

Core model

Each topic can mix multiple sources:

• Web search via query
• RSS/Atom feeds via feeds
• GitHub releases via github_repos

All collected results flow into the same pipeline:

1. gather results
2. deduplicate
3. apply advanced filters
4. score importance
5. classify sentiment
6. alert immediately or save for digest
7. track state and sentiment history

Topic config

Each topic supports these keys:

• id
• name
• query
• keywords
• feeds
• github_repos
• required_keywords
• exclude_keywords
• frequency → hourly|daily|weekly
• importance_threshold → high|medium|low
• channels
• context
• alert_on
• alert_on_sentiment_shift
• ignore_sources
• boost_sources

Example config

{
  "topics": [
    {
      "id": "openclaw-news",
      "name": "OpenClaw Updates",
      "query": "OpenClaw AI assistant update release",
      "keywords": ["OpenClaw", "release", "update"],
      "feeds": ["https://example.com/blog/rss.xml"],
      "github_repos": ["openclaw/openclaw"],
      "required_keywords": ["release"],
      "exclude_keywords": ["rumor", "affiliate"],
      "frequency": "daily",
      "importance_threshold": "medium",
      "channels": ["telegram"],
      "context": "Track product updates and releases",
      "alert_on": ["keyword_exact_match", "github_release"],
      "alert_on_sentiment_shift": true,
      "ignore_sources": [],
      "boost_sources": ["github.com"]
    }
  ]
}

Advanced filters

required_keywords

All listed terms must appear in the title/snippet before scoring.

"required_keywords": ["release", "stable"]

exclude_keywords

Any matching term filters the result out before scoring.

"exclude_keywords": ["beta", "rumor", "affiliate"]

This is intentionally simple boolean logic:

• any exclude match → reject
• any missing required term → reject

RSS/Atom feeds

Direct feeds

python3 scripts/manage_topics.py add "Security Feeds" \
  --feeds "https://example.com/rss.xml,https://example.com/atom.xml" \
  --keywords "security,CVE,patch"

Feed discovery

Discover feeds from a normal website URL:

python3 scripts/manage_topics.py discover-feed https://example.com/blog
python3 scripts/monitor.py --discover-feed https://example.com/blog

Add a topic and auto-discover feeds in one step:

python3 scripts/manage_topics.py add "Vendor Blog" \
  --discover-feeds "https://example.com/blog" \
  --keywords "release,announcement"

OPML import

Import feed subscriptions from an OPML file:

python3 scripts/manage_topics.py import-opml feeds.opml

Imported topics default to daily / medium unless you override:

python3 scripts/manage_topics.py import-opml feeds.opml --frequency hourly --importance high

Feed caching

Feed polling uses feedparser and stores per-feed cache data in monitor state:

• etag
• last-modified
• last check metadata

That allows efficient conditional requests and avoids reprocessing unchanged feeds.

GitHub release monitoring

Track repo releases with:

"github_repos": ["openclaw/openclaw", "anthropics/claude-code"]

These map to GitHub Atom feeds automatically:

• https://github.com/openclaw/openclaw/releases.atom
• https://github.com/anthropics/claude-code/releases.atom

CLI example:

python3 scripts/manage_topics.py add "CLI Releases" \
  --github-repos "openclaw/openclaw,anthropics/claude-code" \
  --keywords "release,version"

GitHub release items are labeled clearly in alerts.

Sentiment analysis

Each scored finding also gets a sentiment label:

• positive
• negative
• neutral
• mixed

Alerts and digest entries include that sentiment.

Sentiment shift alerts

Enable:

"alert_on_sentiment_shift": true

When enabled, a result that changes sentiment versus the topic’s previous sentiment history gets promoted for alerting.

State tracks:

• last_sentiment
• sentiment_history

Scripts

scripts/manage_topics.py

# Add topic
python3 scripts/manage_topics.py add "Topic Name" \
  --query "search query" \
  --keywords "word1,word2" \
  --feeds "https://example.com/rss.xml" \
  --github-repos "openclaw/openclaw" \
  --required-keywords "release" \
  --exclude-keywords "beta,rumor"

# List topics
python3 scripts/manage_topics.py list

# Edit topic
python3 scripts/manage_topics.py edit topic-id --feeds "https://example.com/rss.xml"

# Discover feeds
python3 scripts/manage_topics.py discover-feed https://example.com/blog

# Import OPML
python3 scripts/manage_topics.py import-opml feeds.opml

# Test topic
python3 scripts/manage_topics.py test topic-id

scripts/monitor.py

python3 scripts/monitor.py
python3 scripts/monitor.py --dry-run
python3 scripts/monitor.py --topic openclaw-news --verbose
python3 scripts/monitor.py --discover-feed https://example.com/blog

Alert output

Alerts can now include:

• source label (Web, Feed, GitHub Release)
• score and reason
• sentiment
• sentiment shift marker when applicable

Installation note

Feed support uses Python feedparser.

Install if needed:

pip3 install feedparser

Troubleshooting

Feeds not showing results

• verify the feed URL manually
• try discover-feed against the site URL
• install feedparser
• run python3 scripts/monitor.py --dry-run --verbose

Too much noise

• tighten required_keywords
• add exclude_keywords
• increase importance_threshold

Missing GitHub release alerts

• confirm repo is owner/repo
• verify releases exist on GitHub
• test with --verbose

Notes

• Search and feed results are merged into the same scoring pipeline.
• Existing web-search behavior remains supported.
• Digest entries store sentiment so weekly output reflects tone changes too.