mcpField
A
ai-tech-summary

ai-tech-summary

github.com/tiangong-aiProductivity1168 installsUpdated 1w ago
Repository

About ai-tech-summary

ai-tech-summary is a skill for agents that builds evidence context from RSS records stored in SQLite, then lets the agent write the requested digest. It supports daily, weekly, monthly, and custom date ranges, with selected fields and evidence references.

Key features

  1. 01Daily, weekly, monthly, and custom periods
  2. 02SQLite RSS record retrieval
  3. 03Selectable fields and record limits
  4. 04Evidence anchors using entry IDs and URLs
  5. 05Metadata-only mode when full text is absent

Use cases

  1. 01Preparing a weekly technology news digest
  2. 02Summarizing RSS entries for a custom date range
  3. 03Creating evidence-grounded summaries from retrieved records

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/tiangong-ai/agent-skills --skill ai-tech-summary

After installing, type/ai-tech-summaryin Claude Code to confirm it loaded.

Details

From the project README

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
  1. 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.
  1. 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

Comments

Markdown supported