Claude mem

Skill alebgl77/claude-inc/skills/claude-mem

Persistent filesystem memory across sessions — maintains memory/ with decisions.md, context.md, glossary.md and sessions/YYYY-MM-DD.md, loads and summarizes state at session start, appends decisions and open threads at session end, compacts monthly, and refuses secrets and transient noise. Use when the user says "remember this", "what did we decide about X", "pick up where we left off", or when starting or closing a session on a long-running project.From its SKILL.md

Install

npx -y skills add alebgl77/claude-inc --skill claude-mem

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

8 stars8 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.

SKILL.md

3.8 KB, 823 tokens by cl100k_base, as published. Nobody here has run it

Claude-Mem — Memory Keeper

"Persistent memory"

Chat context evaporates; files don't. Give every project a paper trail the next session can boot from.

When to use

"Pick up where we left off" — session start on any long-running project
"Remember this" / "log that decision" the moment a choice lands
"What did we decide about <topic>, and why?"
"What does <codename/acronym> mean again?" — glossary lookups
"Wrap up" / "save state before we stop" — session end

Workflow

Locate or initialize memory/ at the project root: decisions.md, context.md, glossary.md, sessions/. Create missing files with headers; never overwrite existing ones.
Session start: read context.md, decisions.md, and the last two session files; deliver a ≤10-bullet summary — active decisions, open threads, next actions — and confirm it matches the user's picture.
During work, log decisions the moment they land: append to decisions.md with date, decision, rationale, and alternatives rejected. Same-session or it didn't happen.
New shorthand — nickname, acronym, codename — gets a one-line glossary.md entry on first use.
Session end: write sessions/YYYY-MM-DD.md — what got done, decisions made (pointer, not copy), open threads, next actions.
Monthly compaction: fold session files older than 30 days into context.md (keep decisions, threads, and state; drop the play-by-play), delete the folded files, and dedupe decisions.md keeping the latest ruling per topic.
Refuse to store: secrets, tokens, passwords, keys; personal data beyond names and roles; transient noise (build logs, one-off stack traces, dead ends already resolved). Say what was excluded and why.

Output format

memory/
├── decisions.md       # append-only choices + rationale
├── context.md         # compacted long-term project state
├── glossary.md        # shorthand → meaning
└── sessions/
    └── 2026-07-11.md  # one file per working session

decisions.md entry:
## 2026-07-11 — Postgres over SQLite for prod
Why: concurrent writers. Rejected: SQLite (single-writer lock), DynamoDB (no team experience).

glossary.md entry:
ATLAS — internal codename for the billing rewrite (started 2026-05)

sessions/YYYY-MM-DD.md:
# Session 2026-07-11
Done: <shipped work, one line each>
Decisions: → decisions.md (2026-07-11)
Open threads: <unresolved, with current state>
Next: <first action for the next session>

Quality bar

Session opened with a memory summary, never a cold start
Every decision logged same-session with rationale and rejected alternatives
Session file written before the session closes
Zero secrets, credentials, or transient noise in any memory file
Compaction shrinks bytes without losing a single decision or open thread
Summaries stay ≤10 bullets — memory serves recall, not re-reading

Example

Ask: "We're going with JWT over server sessions — remember that, then wrap up."

Produced: appended ## 2026-07-11 — JWT over server sessions (why: stateless scaling; rejected: session store adds a Redis dependency) to decisions.md, wrote sessions/2026-07-11.md with two open threads (refresh-token rotation, logout semantics), and echoed the entry back for confirmation.

Credits

Inspired by the claude-mem project.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most context ai engineering skills give in 823 tokens

Counted across 1,328 of the 2,349 authors here whose files we hold, read 2026-09-06

Dispatch a fresh subagent for each taskin 76 of 1328, across 59 files
Perform spec compliance review before code quality reviewin 44 of 1328, across 34 files
Dispatch a final code reviewer after all tasksin 38 of 1328, across 26 files
Answer subagent questions before allowing implementationin 36 of 1328, across 26 files
Use the least powerful model capable of the taskin 33 of 1328, across 26 files
Create a TodoWrite list for all tasksin 32 of 1328, across 22 files
Perform a task review after each implementationin 31 of 1328, across 24 files
Extract all tasks and context from the planin 29 of 1328, across 20 files
Provide full task text to subagentsin 28 of 1328, across 20 files
Use git worktrees for isolated workspacesin 25 of 1328, across 20 files
Specify the model explicitly when dispatching a subagentin 23 of 1328, across 18 files
Execute all tasks from the plan without stoppingin 21 of 1328, across 16 files

Said here and by no other author read

Initialize memory directory at project root
Log decisions with rationale and rejected alternatives
Add new shorthand to glossary
Compact session files older than thirty days
Deduplicate decisions monthly
Confirm memory summary with user

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.