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

AI Tech Summary

Core Goal

• Pull the right records and fields for a requested time range.
• Package evidence into a compact JSON context for RAG.
• Let the agent synthesize final summary text from retrieved evidence.
• Support daily, weekly, monthly, and custom time windows.

Triggering Conditions

• Receive requests for daily, weekly, or monthly digests.
• Receive requests for arbitrary date-range summaries.
• Need evidence-grounded summary output from RSS entries/fulltext.
• Need agent-generated summary style rather than rigid scripted report format.

Input Requirements

• Required tables in SQLite: feeds, entries (from ai-tech-rss-fetch).
• Optional table: entry_content (from ai-tech-fulltext-fetch).
• Shared DB path should be the same across all RSS skills.
• In multi-agent runtimes, set AI_RSS_DB_PATH to one absolute DB path for this agent.

RAG Workflow

1. Retrieve evidence context by time window.

export AI_RSS_DB_PATH="/absolute/path/to/workspace-rss-bot/ai_rss.db"

python3 scripts/time_report.py \
  --db "$AI_RSS_DB_PATH" \
  --period weekly \
  --date 2026-02-10 \
  --max-records 120 \
  --max-per-feed 20 \
  --summary-chars 8192 \
  --fulltext-chars 8192 \
  --pretty \
  --output /tmp/ai-tech-weekly-context.json

2. Load retrieval output and generate final summary in agent response.

• Read query, dataset, aggregates, records.
• Prioritize records as evidence source.
• Mention key trends, major events, and notable changes grounded in records.

3. Include evidence anchors in summary.

• Reference entry_id, feed, and URL for key claims.
• If retrieval is truncated, state that summary is based on sampled top records.

Time Window Modes

• --period daily --date YYYY-MM-DD
• --period weekly --date YYYY-MM-DD
• --period monthly --date YYYY-MM-DD
• --period custom --start ... --end ...
• Time filtering is always based on entries.first_seen_at (UTC).

Custom boundaries support both YYYY-MM-DD and ISO datetime.

Field Selection for RAG

• Use --fields to control token budget and relevance.
• Default fields are tuned for summarization:
  • entry_id,timestamp_utc,timestamp_source,feed_title,feed_url,title,url,summary,fulltext_status,fulltext_length,fulltext_excerpt
• Common minimal field set for tight context:
  • entry_id,timestamp_utc,feed_title,title,url,summary

Recommended Agent Output Pattern

• Use this order in final response:
  1. Time range scope
  2. Top themes/trends
  3. Key developments (grouped)
  4. Risks/open questions
  5. Evidence list (entry ids + URLs)

Configurable Parameters

• --db
• AI_RSS_DB_PATH (recommended absolute path in multi-agent runtime)
• --period
• --date
• --start
• --end
• --max-records
• --max-per-feed
• --summary-chars
• --fulltext-chars
• --top-feeds
• --top-keywords
• --fields
• --output
• --pretty
• --fail-on-empty

Error Handling

• Missing feeds/entries: fail fast with setup guidance.
• Invalid date/time/field list: return parse errors.
• Missing entry_content: continue in metadata-only mode.
• Empty retrieval set: return empty context; optionally fail with --fail-on-empty.

References

• references/time-window-rules.md
• references/report-format.md

Assets

• assets/config.example.json

Scripts

• scripts/time_report.py