agentsclimarketplace

Mem sync

Skill akunzai/agent-skills/skills/mem-sync

My personal agent skills for AI coding assistants

Install
npx -y skills add akunzai/agent-skills --skill mem-sync

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

DEPRECATED — Project short-term memory is now stored globally at ~/.agents/memories/projects/<proj-slug>/ without requiring Git sync operations.

SKILL.md

4.5 KB, ~1.1k tokens by cl100k_base, as published. Nobody here has run it

mem-sync — Cross-Device Memory Sync (DEPRECATED)

[!NOTE] mem-sync is deprecated. Project short-term memory has been migrated out of Git repositories to central device-local storage under ~/.agents/memories/projects/<proj-slug>/. Git-based memory synchronization commands are no longer required.

Per-User Branch

The sync branch is derived from git config user.email: the part before @, lowercased and slugified (non-alphanumeric runs become -). [email protected] → memories/user. Each user pushes/pulls only their own branch, so two developers on the same repo never collide on same-date log filenames.

Windows

On Windows, run this skill's script through Git Bash (Git for Windows). A PATH guard at the top of the script prepends the MSYS usr/bin so find/tar are not shadowed by the native same-name tools; if a required POSIX tool is still missing, the script aborts with an explicit error naming the missing tool.

Commands

Discover the script path dynamically (do NOT hardcode):

  1. Global scope: ~/.agents/skills/mem-sync/scripts/mem-sync-git.sh
  2. Project scope: <repo>/skills/mem-sync/scripts/mem-sync-git.sh
  3. Fallback: search for mem-sync-git.sh.
  • mem-sync-git.sh pull — record local WIP, fetch+rebase remote, copy back. Run at session start.
  • mem-sync-git.sh push — same merge, then push. Run at session end or after promoting candidates.
  • mem-sync-git.sh (no argument) — defaults to status.
  • mem-sync-git.sh status — read-only: print the resolved remote/branch header, then summarize how local .memories/ differs (in sync, or local-only / remote-only / modified files).
  • mem-sync-git.sh diff — read-only: like status but prints the full unified diff (local vs remote).
  • mem-sync-git.sh print-branch — print the derived per-user branch and exit (machine-readable, used by tooling/tests).
  • mem-sync-git.sh print-remote — resolve and print just the sync remote (env → auto-detect) and exit (machine-readable). status already shows this; use print-remote when a script needs the bare value. Exits non-zero with guidance if the remote is ambiguous.

See references/git-sync-workflow.md for mechanics, anti-pollution rules, and conflict handling.

Avoid a Race with pull

pull, push, and compact mutate .memories/ — treat them as a critical section. Do not read .memories/ in parallel with those commands.

pull briefly removes and recreates the local .memories/ directory while copying the synchronized snapshot back from the isolated worktree. A read racing against that window — grep, Select-String, Get-Content, cat, or similar — can observe the transient missing-directory state and report a false error. Wait for the sync command to finish successfully before scanning handoffs, candidates, or daily logs.

Remote resolution

The sync remote is resolved in this order:

  1. MEM_SYNC_REMOTE env var — explicit one-off override; highest priority, never persisted.
  2. Auto-detect from the repo's own push configuration (not by remote name):
    • the current branch's push target resolved by Git itself (git for-each-ref --format='%(push:remotename)': branch.<name>.pushRemote → remote.pushDefault → tracking remote) → use that remote, so memory follows wherever the repo actually pushes (e.g. a fork's writable remote), regardless of the name origin;
    • no push target configured but exactly one remote exists → use it;
    • no push target and multiple remotes → ambiguous: the command lists the remotes and exits non-zero. Give the current branch a push target (git push -u <remote> <branch>) or rerun with MEM_SYNC_REMOTE=<name>, then retry.

The remote is recomputed from live Git config on every run, so retargeting your push remote is picked up automatically — there is no stored memory-sync remote to go stale. The per-user branch name (memories/<email-localpart>) is unaffected by the remote choice.

If a command reports an ambiguous remote set, relay the printed remote list to the user and ask which remote to use, then set a push target (or MEM_SYNC_REMOTE=<name>) before retrying.

Anti-Pollution

Never commit .memories/ to an ordinary development branch. If the user asks to "commit memory" or "sync daily notes to Git", intercept and route through this syncer.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.