Claude Memory Canary — Development skill for Claude Code

Memory Canary for Claude Code (free)

One Python file that proves your Claude Code sessions on two machines really share memory. If you sync notes between a desktop and a laptop (a git repo, a shared folder, anything), this is the test that tells you whether the other side actually got them, instead of trusting that the push went out.

It is the "two-facts test" done at the right layer: seed two unguessable facts on machine A, check for them on machine B. PASS or FAIL, with the reason and the age.

Use it

You need Python 3.8+, git, and a git repository both machines clone (keep it private).

1. Prove the script works on this machine (builds a throwaway repo, touches nothing else):

   python memory_canary.py selftest

   Five lines, all PASS.
2. On machine A:

   python memory_canary.py seed --repo ~/memory

   It writes working/canary.md with a code word and a number, commits and pushes, and prints both.
3. On machine B:

   python memory_canary.py check --repo ~/memory

   It pulls, then prints PASS only if the canary is there, unchanged, written by a different machine and under 24 hours old (--max-age-h to change).
4. The human half: open a Claude Code session on machine B and ask, in your own words, what the canary's code word is. A plausible guess instead of the word means your sessions are not loading what arrived.

Exit code 0 = PASS, 1 = FAIL, so it can run from a scheduler or a monitor.

The full kit

Tudlab Code: Memory turns this into a working setup: a session-start pull and stop-hook push of your private memory repo with a merge-never-clobber policy, Claude Code memory notes that travel by tag, lessons loaded by topic the first time a session needs them, a promote job for long-term notes, and a 19-page guide. Write-up of the design: https://tudlab.com/memory US$29: https://gumroad.com/l/tudlab-code-memory

MIT licence.