Imported from tiangong-ai/agent-skills (sustainability-rss-fetch/SKILL.md). Install upstream with npx skills add tiangong-ai/agent-skills --skill sustainability-rss-fetch. Copyright stays with the author.

Sustainability RSS Fetch

Core Goal

• Ingest all RSS/Atom items into SQLite before topic filtering.
• Use doi as the primary key in entries.
• Keep RSS metadata isolated in its own DB file.
• After semantic screening, keep relevant rows and prune non-relevant rows to DOI-only.

Triggering Conditions

• Receive a request to import sustainability feeds and persist all fetched records first.
• Receive a request to do prompt-based topic screening after DB ingestion.
• Receive a request to convert irrelevant rows into lightweight DOI-only records.
• Need stable DOI-keyed storage for downstream API/fulltext/summarization.

Mandatory Workflow

1. Prepare runtime and RSS metadata DB path.

python3 -m pip install feedparser
export SUSTAIN_RSS_DB_PATH="/absolute/path/to/workspace-rss-bot/sustainability_rss.db"
python3 scripts/rss_subscribe.py init-db --db "$SUSTAIN_RSS_DB_PATH"

2. Collect RSS window and ingest all fetched items first.

python3 scripts/rss_subscribe.py collect-window \
  --db "$SUSTAIN_RSS_DB_PATH" \
  --opml assets/journal.opml \
  --start 2026-02-01 \
  --end 2026-02-10 \
  --max-items-per-feed 150 \
  --topic-prompt "筛选与可持续主题相关的文章：生命周期评价、物质流分析、绿色供应链、绿电、绿色设计、减污降碳" \
  --output /tmp/sustainability-candidates.json \
  --pretty

3. Screen candidates in agent context (semantic, not regex-only).

• Use topic_prompt + user instructions.
• Produce selected candidate_id list.

4. Mark selected rows as relevant and prune unselected rows.

python3 scripts/rss_subscribe.py insert-selected \
  --db "$SUSTAIN_RSS_DB_PATH" \
  --candidates /tmp/sustainability-candidates.json \
  --selected-ids 3,7,12,21

Result:

• selected candidates: is_relevant=1, keep metadata.
• unselected candidates: clear metadata fields, keep DOI-only row (is_relevant=0).

Optional Maintenance Sync

python3 scripts/rss_subscribe.py sync --db "$SUSTAIN_RSS_DB_PATH" --max-feeds 20 --max-items-per-feed 100

Source Management

python3 scripts/rss_subscribe.py add-feed --db "$SUSTAIN_RSS_DB_PATH" --url "https://example.com/feed.xml"
python3 scripts/rss_subscribe.py import-opml --db "$SUSTAIN_RSS_DB_PATH" --opml assets/journal.opml

Query Data

python3 scripts/rss_subscribe.py list-feeds --db "$SUSTAIN_RSS_DB_PATH" --limit 50
python3 scripts/rss_subscribe.py list-entries --db "$SUSTAIN_RSS_DB_PATH" --limit 100

Data Contract

• feeds table: subscription and fetch state.
• entries table (doi PK):
  • metadata fields (title/url/summary/categories/...)
  • doi_is_surrogate (when no DOI is present in source)
  • is_relevant (1 relevant, 0 pruned non-relevant, NULL not labeled yet)
• Non-relevant rows are pruned to DOI-only payload for storage efficiency.

Configurable Parameters

• --db
• SUSTAIN_RSS_DB_PATH
• --opml
• --feed-url
• --use-subscribed-feeds
• --topic-prompt
• --start/--end
• --max-feeds
• --max-items-per-feed
• --user-agent
• --cleanup-ttl-days

Error and Boundary Handling

• Feed/network failure: continue other feeds and keep errors in feed state.
• Missing feedparser: return install guidance.
• Missing DOI in RSS item: create deterministic surrogate DOI key to keep full-ingestion guarantee.
• Invalid selected IDs: fail fast before label/prune write.

References

• references/input-model.md
• references/output-rules.md
• references/time-range-rules.md

Assets

• assets/journal.opml
• assets/config.example.json

Scripts

• scripts/rss_subscribe.py