Release notes · 0.2 · 7 October 2026

Actualis now reads Copilot CLI. Post your card.

Actualis reads the transcripts your coding agents already write and reports what actually ran. 0.2 adds GitHub Copilot CLI alongside Claude Code and Codex, and a command that turns your own numbers into an image you can post.

An Actualis card: 76% of my agents' shell commands ran with nobody approving them, over the last 31 days

$ uv tool install actualis
$ actualis --card

What’s new

GitHub Copilot CLI is read

Shell commands go into the same audit as Claude Code and Codex, with cost per model, which commands were approved and which ran unprompted, refusals, and premium requests. --agent copilot reads it alone.

`actualis --card`

A 1200×630 SVG and PNG of your own numbers: how many shell commands ran with nobody approving them, what your agents cost at list price, or how many commands they ran. Two styles, hero and terminal.

Nothing identifying on the card

Counts, four command categories and public model names only. No project, branch, path, command or credential can reach it, and the test suite checks the SVG, the PNG pixels and the caption for each of them.

One definition of “unsupervised”

The card, --share, the MCP server and the OWASP AISVS mapping now agree on what ran without a human approving it.

0.2.1 — The GitHub Action loads again: a comment in 0.2.0's action.yml held an expression GitHub refused to parse. The CLI was not affected.

Ready to paste. The link at the end of each post unfurls with the card above.

LinkedIn, in your own voice

I just shipped Actualis 0.2.

Actualis is a free, local tool that reads the logs your coding agents already write and tells you what actually ran: which shell commands, which credentials passed through them, and what it all cost.

0.2 adds two things:

→ GitHub Copilot CLI support, alongside Claude Code and Codex.
→ actualis --card, which turns your own numbers into an image you can post.

The first thing I checked was my own Copilot history. More than 9 in 10 of its shell commands ran without anyone approving them. Nothing went wrong; that is simply what "auto" means, and I had stopped noticing.

The card shows that number for your machine. It carries counts and public model names only: no project names, paths, commands or credentials, and the test suite checks every image for them.

It all runs locally. No account, no telemetry, no network calls.

uv tool install actualis
actualis --card

What's your number?

https://actualis.app/releases/0-2/

LinkedIn, as Digital Foundry

Actualis 0.2 is out.

Actualis reads the transcripts coding agents already write (Claude Code, Codex, and now GitHub Copilot CLI) and reports what actually ran: shell commands, exposed credentials, supervision and cost. It runs locally and read-only, with no network calls.

New in 0.2:
• Copilot CLI sessions: commands, cost per model, which commands were approved and which ran unprompted, and refusals.
• actualis --card: a shareable 1200×630 image of your own numbers (unsupervised share, spend or command volume) with nothing identifying on it.
• One definition of "unsupervised" across every report.

Free and open source under AGPL-3.0.

uv tool install actualis
https://actualis.app/releases/0-2/

Short, for X, Bluesky or a newsletter

Actualis 0.2: now reads GitHub Copilot CLI, and actualis --card turns what your coding agents actually ran into a postable image. Local, read-only, no network.

uv tool install actualis
https://actualis.app/releases/0-2/

Full changelog on GitHub · All releases