Skip to content

docs: add a one-minute terminal demo to the README #310

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Defilan merged 1 commit into from
Nov 2, 2026
Merged

docs: add a one-minute terminal demo to the README #310

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 10 additions & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,16 @@ model's behavior: see what each result means, and the detection ceiling, below.

*Socair* (SUH-ker) is Irish for "at ease, settled, secure".

<p align="center">
<img alt="A terminal session. A repository shaped like the October 2026 Hugging Face miner campaign is scanned: promotion withheld, with a Remote code LEAD naming the module its config tells the loader to import. Qwen/Qwen3-0.6B is pulled through the airlock at a pinned commit and scanned: every row PASS, promotion authorized. Both reports are signed with an operator key. The airlock refuses the first, promotes the second into the clean store, and the attestation verifies offline." src="docs/demo/intake.gif" width="820">
</p>

*Two models through the same intake. The withheld one is the detection
benchmark's defanged copy of the campaign's layout: its stand-in payload only
builds a string. Recorded at v0.1.0-rc.8 with
[scripts/record-demo.sh](scripts/record-demo.sh); the 1.5 GB download is
shortened.*

## What it checks

Each check returns **PASS**, **FAIL**, **LEAD**, or **NOT_TESTED**:
Expand Down
Binary file added docs/demo/intake.gif
View file
Open in desktop
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
112 changes: 112 additions & 0 deletions scripts/record-demo.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,112 @@
#!/usr/bin/env bash
# Record the README's terminal demo as a GIF.
#
# scripts/record-demo.sh [<out.gif>] (default docs/demo/intake.gif)
#
# Two models go through the same intake:
# qwen2-tiny the detection benchmark's dir-remote-code-tiny-qwen2
# case, shaped like the October 2026 Hugging Face miner
# campaign: a repo posing as a tiny Qwen2 checkpoint whose
# auto_map names a module that runs at import. Defanged:
# its stand-in payload only builds a string. The benchmark
# test writes it (SOCAIR_BENCHMARK_CORPUS), so the demo
# scans the benchmark's own bytes.
# Qwen/Qwen3-0.6B pulled through the airlock at a pinned commit (about
# 1.5 GB, the only network use).
# Both reports are signed with a fresh demo key; the first is refused at
# promotion, the second is promoted and verified.
#
# Needs go, jq, asciinema 3, and agg. Everything else lands in a temp dir.
# The recording caps idle time at 4 seconds, so the download is shortened.
# SOCAIR_DEMO_FONT_DIR points agg at a font directory (the GIF in the repo
# uses JetBrains Mono).
set -euo pipefail

revision=c1899de289a04d12100db370d81485cdf75e47ca
digest=fbb2e51d2156914d13091c694e8ac226bedc3022c587f31f72ca8831304fb199

# The recorded commands are shown as typed, $D included; eval expands them.
# shellcheck disable=SC2016
if [ "${1:-}" = "--play" ]; then
# The recorded session: runs in the work dir, with socair on PATH.
say() {
printf '\033[90m# %s\033[0m\n' "$@"
sleep 1.5
}
run() {
local cmd=$1 pause=${2:-1.5} i
printf '\033[32m$\033[0m '
for ((i = 0; i < ${#cmd}; i++)); do
printf '%s' "${cmd:i:1}"
sleep 0.015
done
sleep 0.5
printf '\n'
eval "$cmd" || true
sleep "$pause"
}
cls() { printf '\033[2J\033[H'; }

cls
say "A repo shaped like the October 2026 Hugging Face miner campaign: it poses as a tiny" \
"Qwen2 checkpoint, and its config tells the loader to import its Python. Defanged, from" \
"Socair's detection benchmark."
run 'ls qwen2-tiny'
run 'socair scan qwen2-tiny > tiny.json' 3
run "jq -r '.checks[] | select(.status == \"LEAD\") | .evidence' tiny.json" 4

cls
say "A real model, pulled through the airlock at a pinned commit."
run "socair airlock pull --repo Qwen/Qwen3-0.6B --revision $revision" 3
run "D=$digest" 0.5
run 'SOCAIR_PROVENANCE=airlock/incoming/$D/provenance.json socair scan airlock/incoming/$D/Qwen3-0.6B > qwen.json' 0.5
run "jq -r '(.checks[] | \"\\(.status) \\(.name)\"), \"promotion: \\(.promotion_authorization.state)\"' qwen.json" 4

cls
say "Sign both reports with your own key, and trust it at the airlock."
run 'socair key gen --out operator --issuer "Acme ML Platform" && socair airlock trust add operator.pub' 1
run 'socair sign --key operator.key --report tiny.json && socair sign --key operator.key --report qwen.json' 2

cls
say "Promote both. Only one crosses into the clean store, and anyone can verify it offline."
run 'socair airlock promote qwen2-tiny --attestation tiny.dsse.json' 4
run 'socair airlock promote airlock/incoming/$D/Qwen3-0.6B --attestation qwen.dsse.json' 2
run 'socair verify qwen.dsse.json --trusted operator.pub --artifact airlock/clean/$D/Qwen3-0.6B' 3
exit 0
fi

root=$(cd "$(dirname "$0")/.." && pwd)
out=${1:-$root/docs/demo/intake.gif}
case "$out" in /*) ;; *) out=$PWD/$out ;; esac
for tool in go jq asciinema agg; do
command -v "$tool" >/dev/null || { echo "record-demo: $tool is required" >&2; exit 1; }
done

work=$(mktemp -d)
trap 'rm -rf "$work"' EXIT
mkdir -p "$work/bin" "$work/demo"
(cd "$root" && go build -o "$work/bin/socair" ./cmd/socair)
(cd "$root" && SOCAIR_BENCHMARK_CORPUS="$work/corpus" go test ./internal/benchmark -run '^TestDetectionBenchmark$' -count=1 >/dev/null)
cp -R "$work/corpus/dir-remote-code-tiny-qwen2/model" "$work/demo/qwen2-tiny"

cd "$work/demo"
export PATH="$work/bin:$PATH" SOCAIR_STORE=airlock
socair airlock init >/dev/null
asciinema rec --quiet --headless --window-size 100x26 --idle-time-limit 4 \
-c "bash $root/scripts/record-demo.sh --play" "$work/demo.cast"

# The demo must show what its caption says. The cast's output events, joined,
# are the session's output stream.
tail -n +2 "$work/demo.cast" | jq -j 'select(.[1] == "o") | .[2]' > "$work/demo.out"
for want in 'promotion: withheld' 'imported with trust_remote_code' 'promotion: authorized' \
'promotion refused' "promoted $digest" 'verified: issued by Acme ML Platform'; do
grep -qF "$want" "$work/demo.out" || { echo "record-demo: the recording does not show: $want" >&2; exit 1; }
done

theme=0f2333,e7eef1,0b1a26,f2877a,4cd3ae,e9b65a,7fb3e0,c9a0dc,7fd6d0,e7eef1,8fa3af,f6a499,6ee0c0,f0c97e,9cc6ea,d8b8e6,9fe3de,f4f7f8
font_args=(--font-family "JetBrains Mono,Menlo,DejaVu Sans Mono")
[ -n "${SOCAIR_DEMO_FONT_DIR:-}" ] && font_args+=(--font-dir "$SOCAIR_DEMO_FONT_DIR")
mkdir -p "$(dirname "$out")"
agg --quiet --theme "$theme" "${font_args[@]}" --font-size 16 --idle-time-limit 4 --last-frame-duration 8 \
"$work/demo.cast" "$out"
echo "wrote $out"
Loading