Check what really happened.

Eight small, open-source command-line tools. Each one takes something that is usually trusted on its word (a backup, a scheduled job, a lockfile, an AI agent's edit, an LLM's answer) and checks it against reality. MIT licensed, on PyPI, tested on Linux, macOS and Windows.

Data and scheduled jobs

firedrill

Restores PostgreSQL, MySQL and MariaDB backups into a throwaway, version-matched container, checks SQLite files with no container at all, and proves the result is usable.

CRITICAL ARCHIVE_TRUNCATED
  the plain-SQL dump doesn't end with pg_dump's
  'dump complete' line: it was cut off
HIGH     SEQUENCE_BEHIND
  sequence public.customer_id_seq is at 1,
  behind the largest value of 1774

complete backup: PASS   cut-off backup: FAIL
pip install firedrill docs

escrow

A dead man's switch for cron jobs, GitHub Actions schedules and systemd timers. Flags silence, not just failure.

[!!] 'nightly-backup' missed its run due
     2026-11-18 01:00 UTC (0 1 * * *, 30m grace);
     last pinged 1d ago
[??] 'weekly-report' has never pinged in
pip install escrow-evidence

Code and supply chain

carabiner

Makes a repository secure by default, ratchets so only new findings fail the build, and proves the controls actually fire.

CRITICAL CI001  .github/workflows/pr.yml
  job 'hello' runs on pull_request_target and
  checks out the PR head -- untrusted code runs
  with your secrets

2 new, 340 accepted (carabiner debt)
pip install carabiner-sec docs

lockstep

Does what's installed right now match the lockfile? Python and npm, checked inside the running environment.

[!=] 'click' is pinned to 8.1.8 but 8.1.7 is installed
[--] 'requests' is pinned to 2.32.3 but is not installed
[++] 'six' 1.16.0 is installed but not declared anywhere
pip install lockstep-evidence

AI agents and LLM output

receipt

A tamper-evident receipt for what a command or an AI coding agent (Claude Code, Codex, Cursor) actually changed on disk.

$ receipt run --declare app/auth.py -- python fix.py
[FAIL] touched 1 undeclared file(s): app/payments.py
pip install receipt-evidence

sourced

Checks an LLM's answer against its own sources, claim by claim. Measured on RAGTruth's 20,261 test sentences.

[OK] Microsoft reported revenue of $56 billion.
[XX] Microsoft grew 40% year over year.
       claims 40% near 'Microsoft', but the source's own
       text near that entity says ['56 billion', '8%']
pip install sourced-evidence

Devices and reading

drift

Does a device's real network traffic match what it's allowed to contact? Checked against a packet capture or Zeek logs.

[XX] 145.254.160.237: 'test workstation' contacted domain(s)
     outside its declared allow-list: pagead2.googlesyndication.com
pip install drift-evidence

satchel

A local-first reading archive: saves the article text, not just the link, with offline full-text search and a browser extension.

$ satchel add https://example.com/some-article
saved #4: Why Local-First Software Is Worth the Extra Effort

$ satchel search "FTS5"
#4    Why Local-First Software Is Worth the Extra Effort
pip install satchel-reader