Best way to track earnings season in Claude Code

One call for the whole month

The calendar takes a date range and hands back every company reporting inside it. Each row carries the symbol, the date, and — this is the part that makes it useful rather than decorative — the expected EPS and revenue alongside the actual figures, which sit empty until the company reports.

Three of those four 13 October names are large banks, which is not a coincidence — banks cluster at the start of the season. Seeing it laid out is the point: a calendar is how you find out that several of the companies you follow land on the same date, which is a planning fact you would rather have in September than on the day. The response carries the date, not a time of day, so how they are spaced within the day is a question it does not answer.

The same call, after the print

The reason to use the calendar rather than a list of dates you typed yourself is what happens next. After a company reports, the actual EPS and revenue appear in the same record next to the estimate that was already there. So the before-and-after comparison is not two sources and a join — it is the same call, run again.

That makes the update trivially cheap, which matters more than it sounds. A workflow where refreshing costs one call and a few seconds gets refreshed. A workflow where it means rebuilding a spreadsheet gets refreshed in theory.

There is a second reason the calendar beats a list of dates you maintain yourself: dates move. Companies bring a report forward or push it back, and a hand-kept list goes quietly wrong rather than loudly wrong — you find out by missing the print. Because the call is cheap enough to re-run whenever you are curious, the schedule you are working from is the schedule as it stands, not the schedule as it stood when you built the table.

That also makes the calendar a reasonable first call before any deeper work. Knowing that a company reports in four days changes what is worth doing now: pulling three years of statements the day before a print is better timed than pulling them the week after, when the numbers you just paid for are about to be superseded.

Turning it into a routine

1. Pull the window once, early

   A month ahead is enough to see the clusters and notice the days when three of your companies report together. One call covers the whole range.

2. Re-run on the morning

   The same call, the same range, and the actual figures are now populated for anything that has reported. There is nothing to reconcile because nothing moved.

3. Go deeper only where it matters

   For a company whose print you care about, follow up with the income statement and ratios — a few more calls, and only for the handful that earned the attention.

4. Keep each run's file

   Saving the response per run gives you the estimate as it stood before the print, which is the thing that quietly gets revised and then forgotten.

What it will not do for you

It will not tell you whether a beat is a good result. Expectations drift in the weeks before a print, a company can beat a lowered bar, and none of that context lives in a calendar row. The call is a scheduler and a scoreboard; reading the game is still your job, and the data does not pretend otherwise.

It also covers what it covers. The window returns the companies the service tracks for those dates, so treat a name's absence as a question rather than as evidence it is not reporting — the same way you would with any single source.

Set it up once

npm install --global @actionway/cli@latest
actionway init
mkdir -p "$HOME/.claude/skills"
cp -R "$HOME/.actionway/skill/actionway" "$HOME/.claude/skills/actionway"

You need Node.js 20 or newer and an Actionway account. The skills roundup lists every financial call and what each one answers, and the peer comparison guide covers putting several companies on the same fields once the results are in. Codex users copy the same folder into ~/.codex/skills/actionway instead, and the get started page lists the steps per client.