session-memory
Skill32 starsChanged 15 days agoMITView on GitHub
Write what the session produces into your vault AS IT HAPPENS, not at the end: a decision taken, a tool, scheduled task, path or running service created or changed, a measurement that contradicts a note, something left to do, and the closing checklist. Not for lessons that should improve a skill (task-observer owns those, rule R130); not for reloading context at session start (boot skill).
12 other projects use a version of this.
Adopted by
Most recently taken up by
- apappascs/claude-code-sessionstheir own version, first seen Nov 1, 2026
- alexeyban/EKOStheir own version, first seen Nov 1, 2026
- InanisV/my-claude-skillstheir own version, first seen Aug 22, 2026
- Caelumca/claude-codex-skeletontheir own version, first seen Aug 18, 2026
- KogaCyber/jarvis-skillstheir own version, first seen Aug 17, 2026
- moriwo-dev-ai/amateras-registrytheir own version, first seen Jul 26, 2026
Other versions12 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 12
- oaustegard/claude-skills★ 158same name
- alexeyban/EKOS★ 26same name
- DangQuangSE/team-development-skills★ 19same name
- lukasreese/claude-skills-starter-kit★ 3same name
- apappascs/claude-code-sessions★ 2same name
- InanisV/my-claude-skills★ 1same name
- Caelumca/claude-codex-skeleton★ 0same name
- ieteerapat/ClaudeSkills1.0★ 0same name
- KogaCyber/jarvis-skills★ 0same name
- Mesteriis/Engineering-Bible-AI★ 0same name
2 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository20 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- open-source-jp-scraperSkill
- vault-token-optimizerSkill
- agent-config-auditSkill
- claude-code-routerSkill
- code-correctorSkill
- consolidate-memorySkill
- dev-protocolSkill
- docmancerSkill
- doc-to-skillSkill
- god-modeSkill
- rtkSkill
- skill-creatorSkill
- skill-managerSkill
- subtitle-fixSkill
- tk-bootSkill
- tk-grillSkill
- tk-orchestratorSkill
- token-saviorSkill
The file
We split this skill into the rules it gives an agent
1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.
0
Common
Most of the 11 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 81
Never write a secret, not even described.
“Common” is measured: kept by at least 6 in 10 of the 11 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Session memory — never learn the same thing twice
A session that is cut off records nothing. Write each fact when it happens, into the channels the next session and the human actually read — never into a new parallel file: a lesson filed next to the register is a lesson lost.
Configure once — your paths
Edit this table for your own vault. The defaults assume an Obsidian-style vault.
| Channel | Default path |
| Journal (one dated entry per notable action) | 00_SYSTEM/JOURNAL/entrees/<YYYY>/<MM>/<YYYY-MM-DD>_<project>_<slug>.md |
| Recovery note (state, what runs, what is left) | 00_SYSTEM/05_Hot_Cache/HOT_CACHE.md |
| Tools registry | 80_OUTILS/INDEX.md |
| Automations registry (scheduled tasks, timers) | 00_SYSTEM/AUTOMATIONS_INDEX.md |
| Paths and access routes | 00_SYSTEM/ARBORESCENCE_systeme.md |
| Infrastructure (what actually runs) | Infrastructure_Hub/INDEX.md |
| Coordination board (multi-agent setups) | 00_coordination/ACTIVE_WORK.md + outbox/ of your hub repository |
When to write — without being asked
| Trigger | Immediate write |
| A decision is taken (architecture, threshold, port, frequency) | journal entry + recovery note, with the measurement that settled it |
| A tool is created or changed | tools registry + its own page |
| A scheduled task is added, disabled or removed | automations registry, dated line |
| A folder, path or access route appears | paths registry |
| What runs changes (service, container, port) | infrastructure registry, dated line |
| A piece of work starts or ends (multi-agent) | coordination board: declare before acting, report when done |
| A measurement contradicts a note | dated addition to that note — never rewrite the old line |
| Something is left to do | checkbox in the recovery note |
| End of session | the checklist below |
Not here — task-observer (rule R130). An error, a user correction or a repeated task that
should change a skill is written as a task-observer observation. The journal line only cites it:
see observation 0005. Never copy the observation into the journal.
Journal entry format
---
date: "YYYY-MM-DD"
heure: "HH:MM"
projet: "<project>"
theme: "<theme>"
type: "<decision | livraison | reparation | installation>"
statut: "<fait | en-attente | en-observation>"
titre: "<one line>"
preuve: "<the measurement, commit or file that proves it>"
---
# <title>
- What was done, measured and verified (with dates).
- Rollback: <how to undo it>.
- Left to do: <if any>.
- Lesson for a skill: see observation NNNN (task-observer), if any.Ground rules
- No-drop: archive, never delete. An empty field beats a guessed one.
- Nothing exists until it is written down and dated.
- A measurement beats a note, even a recent one: correct by dated addition, so one can read what was believed and when it stopped being believed.
- A page written by reading code describes an intention, not a state: every tool or
automation page carries a "state VERIFIED on
<date>" line with the measurement behind it. - A script returns 0 only if its announced effect happened.
- Never write a secret, not even described. Scan before copying anything to another host.
Before closing a session
- Are the session's decisions in the journal, each with its proof?
- Does the recovery note say what runs now and what is left?
- Are the registries touched today up to date (tools, automations, paths, infrastructure, coordination board)?
- Did a measurement contradict the vault? Is the dated addition made?
- Were the skill lessons logged as task-observer observations, and cited (not copied)?
- Does the closing answer state the measured backup status?
What is not written down will be paid for again — in time, next session.
Changelog
v1.3 — 2026-11-01
- Public generic version. Role split with task-observer (R130): skill-improvement lessons move to task-observer observations; this skill keeps the journal of facts and decisions and cites observation numbers.
v1.0–v1.2 — 2026-09-14 → 2026-10-21
- Private versions (French,
memoire-de-session).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.