SuperAPI AI Directory

earnings-watch

Data & analytics

View repository

Updated 10-16 00:01

About this skill

Earnings Watch

Runs a weekly sweep and, during active reporting windows, a daily diff. It flags earnings, guidance, and related company announcements for the user's watchlist.

This skill is orchestration only. For analysis method, install LLMQuant/skills.

Model baseline

Use DeepSeek as the default Hermes model. Keep the weekly sweep terse on DeepSeek V4 Flash. Switch to a stronger DeepSeek model only for a longer post-earnings memo.

What it does

  1. Weekly sweep: use news_browse with events=["earnings","guidance"] and the watchlist, then scan filing metadata for names that may be in a reporting window.
  2. Daily active-window diff: re-check names already flagged as active. Read only genuinely new filings or announcements.
  3. Delivery: send a short list of names, dates, links, and next actions.

MCP tools used

Tool Purpose Credit / call
news_browse Earnings/guidance company news by ticker/date 2
sec_filing_browse Recent 8-K/10-Q filing metadata 0
sec_filing_read Read newly detected filing text 1
wiki_search Optional one-time primer for unfamiliar names 1

Cron

# Weekly sweep
hermes cron create "0 18 * * 0" "Run earnings-watch weekly: find watchlist earnings and guidance news, scan filing metadata, and list names that need daily follow-up this week." \
  --skill earnings-watch \
  --name "earnings-watch-weekly" \
  --deliver telegram

# Daily diff for active reporting windows
hermes cron create "0 7 * * 1-5" "Run earnings-watch daily diff only for names currently in an active reporting window. Summarize new company news or newly filed text; stay silent if nothing changed." \
  --skill earnings-watch \
  --name "earnings-watch-daily-diff" \
  --deliver telegram

On some Hermes versions (as of v0.18.x), cron expressions evaluate in host-local time. Confirm next-run time after scheduling.

Cost of watching

Keep runs lean: check remaining credits before heavy calls, prefer a small watchlist, and skip optional primers and filing reads when credits are low. Keep the daily diff to genuinely active names so it stays quiet off-season.

Output contract

For each name, include:

  • ticker;
  • why it was flagged: earnings, guidance, filing, or other company update;
  • published_at or filing date;
  • source link when news_browse provides one;
  • whether the agent recommends a follow-up read, without making an investment recommendation.

Known gap

LLMQuant Data does not expose a certified forward earnings calendar. This skill uses recent company news and filing signals. State that plainly in the message.

Customize

  • Watchlist: edit watchlist.tickers in workspace AGENTS.md.
  • Cadence: keep weekly-only outside earnings season.
  • Date range: use a short date window for daily diffs to avoid old matches.
  • Depth: read filings only when the news/filer metadata shows a real change.