agentleFS
jsmastery-projsmastery-pro/jsm-hq/.agents/skills/remember/SKILL.md

remember

Skill47 starsChanged 33 days agoMITView on GitHub

Automatically restores context at the start of every new session so the agent knows exactly where the project stands. Also saves session state manually at the end when the developer is ready. /remember restore triggers automatically when a new session starts. /remember save is always run manually by the developer before closing a session.

  • Reads credentials

70 other projects use a version of this, 3 of them edited.

Adopted by

Most recently taken up by

Used with

Projects with it also use agent-context-isolation, agentica-infrastructure and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. 0 are common, 12 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 62 edited versions we read keep these.

  • 12

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Never persist secrets. If any sensitive value appears in the conversation, do not write it to any file.

    line 58
  2. Distinctive:

    If unsure whether something is sensitive, treat it as sensitive.

    line 66
  3. Distinctive:

    Stop here. Do not proceed.

    line 90
  4. Distinctive:

    Do not read any other files at this stage.

    line 98
  5. Very specific:

    Read only the most recent entry from context/dev-log/[feature-concern].md — the entry at the top.

    line 113
  6. Distinctive:

    Do not start building. Show the developer what was restored and wait for confirmation.

    line 117
  7. Distinctive:

    Only after confirmation does the session continue.

    line 131
  8. Very specific:

    If context/index.md exists but the dev-log is missing or unclear:

    line 135
  9. Distinctive:

    Do not guess. Surface the gap.

    line 144
  10. Distinctive:

    Infer first: Read the session. If it is obvious what was worked on — use that name without asking.

    line 154
  11. Distinctive:

    Create if it does not exist.

    line 195
  12. Distinctive:

    Update to reflect current project state.

    line 221
  13. Distinctive:

    Keep it under 20 lines. Every line costs tokens on every session.

    line 223
  14. Distinctive:

    Update only lines that changed.

    line 246

“Common” is measured: kept by at least 6 in 10 of the 62 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Project Context File

This skill reads and writes the project context file. The name depends on your tool:

  • Claude Code → CLAUDE.md
  • Cursor → .cursor/rules/rules.md
  • Codex → AGENTS.md
  • Windsurf → .windsurfrules
  • Other tools → check your tool's documentation

Throughout this skill, "project context file" refers to whichever file your tool uses.


AI has no memory between sessions. Every new context window starts blank. This skill fixes that.

Restore runs automatically at the start of every session — the developer never has to think about it. Save is always manual — the developer decides when they are satisfied, when tests pass, when it is a clean stopping point. Never save automatically.


The Rule

Session start → restore runs automatically. Session end → developer runs /remember save when ready.

Save is never automatic. The developer must consciously decide this is a good stopping point before saving. Saving a broken or incomplete state creates a worse starting point than no state at all.


File Structure

This skill reads and writes two places only:

context/
├── index.md                        ← always-on, <20 lines, every session reads this
└── dev-log/
    ├── auth-backend.md             ← on-demand, per feature-concern
    ├── auth-frontend.md
    └── dashboard-ui.md

context/index.md is the always-on file. Under 20 lines always. It tells any new session the current project state without loading anything else. Every save updates it.

context/dev-log/[feature-concern].md is per-session detail. One file per feature-concern. Append-only — never overwritten. Two developers on different features write to different files. No conflicts.


Security Boundary

Never persist secrets. If any sensitive value appears in the conversation, do not write it to any file.

Sensitive data includes:

  • API keys, access tokens, refresh tokens, session tokens
  • Passwords, passphrases, private keys, certificates
  • Cookies, auth headers, connection strings, webhook secrets

Store a redacted placeholder if the detail is useful: [REDACTED_API_KEY]. If unsure whether something is sensitive, treat it as sensitive.


Auto-Restore — Runs at Session Start

This triggers automatically when a new session begins. The developer does not invoke it manually.

Step 1 — Check for context/index.md

Look for context/index.md.

If it does not exist:

No context found.

This looks like a fresh start. If this is a new project:
→ Run /mvp to define what you are building
→ Run /architect for technical decisions

If context should exist, check that the context/ folder
is in the project root.

Stop here. Do not proceed.

If it exists — read it fully. Then read the tool context file:

  • project context file — Claude Code
  • AGENTS.md — Codex
  • .cursor/rules/ — Cursor
  • .windsurfrules — Windsurf

Do not read any other files at this stage.

Step 2 — Identify active work

From context/index.md, identify which feature-concern is active.

Infer first. One feature in progress → restore it without asking.

Ask when ambiguous. Multiple features in progress → ask one question:

Which feature are you picking up?
Active: [list from index.md]

Read only the most recent entry from context/dev-log/[feature-concern].md — the entry at the top. Not the full history.

Step 3 — Confirm before continuing

Do not start building. Show the developer what was restored and wait for confirmation.

Context restored.

Last session: [what was built — specific files and features]
Current state: [what works, what is partial, what is broken]
Decisions locked: [key decisions future work depends on]
Next up: [the specific first action this session]

Is this correct? Confirm to continue, or correct
anything before we start.

Only after confirmation does the session continue.

Step 4 — If context is incomplete

If context/index.md exists but the dev-log is missing or unclear:

I found context/index.md but the dev-log for [feature-concern]
is [missing / unclear about: what is unclear].

Continue with what we have, or fill the gaps first?

Do not guess. Surface the gap. Let the developer decide.


Manual Save — /remember save

The developer runs this when they are satisfied with the session. Not before. Not automatically.

Step 1 — Identify the feature-concern

Infer first. Read the session. If it is obvious what was worked on — use that name without asking.

Ask only when ambiguous:

What feature and concern should I save this as?
Example: auth-backend, auth-frontend, dashboard-ui, payments-api

Step 2 — Extract what matters

Review the session and extract only what a developer needs to continue in a completely fresh context. Not a transcript. The essential state only.

What was built — specific files created or modified. Not "built the auth flow" — "created app/(auth)/login/page.tsx and middleware.ts. OAuth with Google and GitHub working end to end."

Decisions made — choices that are hard to reverse or that future work depends on.

Problems solved — anything that took significant time. So the next session does not solve it again.

Current state — exactly where things stand. What works, what is partial, what is broken.

Next session starts with — the very first specific action. Concrete enough to start immediately.

Open questions — anything unresolved that needs addressing next session.

What not to capture:

  • Implementation details visible in the code
  • Decisions already in project context file or context/architecture.md
  • Anything inferable by reading the codebase
  • Secrets or credential-like values

Step 3 — Security check

Scan everything before writing. Remove or redact any credential, key, token, or secret.

Step 4 — Write the dev-log entry

Append-only. Never overwrite. Add new entry at the top of the file.

File: context/dev-log/[feature-concern].md

Create if it does not exist. Prepend above previous entries if it does.

## Session — [date and time]

### What was built
[Specific files and features completed this session]

### Decisions made
[Decisions that future work depends on]

### Problems solved
[Issues resolved — so they are not solved again]

### Current state
[What works, what is partial, what is broken]

### Next session starts with
[The very first specific action next session]

### Open questions
[Anything unresolved]

Step 5 — Update context/index.md

Update to reflect current project state. Create if it does not exist.

Keep it under 20 lines. Every line costs tokens on every session.

# Project Index

Updated: [date] | Stack: [main stack]

## Active Work
- [feature-concern]: [status, one line] → context/dev-log/[feature-concern].md
- [feature-concern]: [status, one line] → context/dev-log/[feature-concern].md

## Last Decision
[Most recent important decision, one line]

## What Not To Touch
[Locked areas, active constraints, things currently being changed]

## Reference if needed
- Product scope: context/mvp.md
- Architecture: context/architecture.md
- Feature plans: context/plans/

Update only lines that changed. Do not rewrite sections that are still accurate.

Step 6 — Confirm

Saved → context/dev-log/[feature-concern].md
Updated → context/index.md

Close the session when ready. Next session restores automatically.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.