Journal Lite: project journal on a token budget

    1

    A ≤40-line project snapshot and ≤20-line session entries in plain Markdown inside your project. Recall past work by task tag without loading the whole history. No dependencies, no network.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Journal Lite: project journal on a token budget

    Example session with this skill installed

    We're picking up the log parser again. What did we decide last time, and where did we stop?

    • Read your context and instructions
    • Compiled the journal lite

    Summary

    Recalled the parser task from the project journal: read the 40-line snapshot and the 3 newest #parser entries (about 700 tokens), without opening the rest of the history.

    Where we stopped

    Last session2026-11-14, "Switched log parser to regex"
    DecidedRegex instead of json: the nginx logs are not valid JSON
    Donesrc/parse.py, tests/test_parse.py; commit 3f2a1c
    Dead endStreaming parser: 3× slower on 1 GB files, dropped
    NextHandle multi-line stack traces

    Project snapshot (from STATE.md)

    • Goal: CLI that turns nginx access logs into a daily CSV report.
    • In progress: multi-line stack traces in error logs.
    • Open question: keep Python 3.9 support or move to 3.11?
    • Key files: src/parse.py (parser), src/report.py (CSV), tests/.

    Commands used

    head -n 40 .journal/STATE.md; tail -n 5 .journal/INDEX.md
    T="parser"; N=3; awk -v t="$T" -v n="$N" '/^## /{...}' .journal/log/*.md
    

    Next steps

    1. Add a failing test with a 3-line Java stack trace in tests/test_parse.py.
    2. Extend the regex to join continuation lines that start with whitespace.
    3. At wrap-up, record a new #parser entry and update STATE.md.
    Project snapshot (from STATE.md

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Most memory skills record everything and pay for it in context. Journal Lite keeps a short, human-readable journal inside your project and caps what it costs.

    What it does

    • Keeps .journal/STATE.md: a ≤ 40-line snapshot of the project (goal, done, in progress, open questions, next steps).
    • Records one ≤ 20-line entry per session at wrap-up, in a single shell call.
    • Recalls a task by tag: returns only the 3 newest matching entries, never the whole log.
    • Keeps small helper scripts and data files (≤ 1 MB) next to the journal.

    Token budget

    MomentCost
    Start of work≤ 40 + 5 lines, ~800 tokens worst case
    Recording an entry~150–300 tokens
    Recalling a task3 entries

    Limits are enforced inside the shell commands (head, tail, awk), not left to the model's judgment.

    Safe by design

    • Plain Markdown in <project>/.journal/, with its own .gitignore, so it is never committed.
    • Writes only inside the project directory. No network, no dependencies, nothing to install.

    Want it automatic?

    The free, open-source Claude Code plugin loads the snapshot at session start with a single hook and enforces the limits in a tiny Python script: https://github.com/IvanStrazov/claude-journal

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Productivity.

    Needs a POSIX shell (bash or zsh; Git Bash on Windows). The snapshot is read when the agent starts work in a project, not loaded automatically at session start. Entries are cut at 20 lines and the snapshot at 40. Files over 1 MB are not copied into the journal.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed today

    • Free with an Agensi account

    Software
    Bash
    orzsh shell

    Needs a POSIX shell (bash/zsh). On Windows, use Git Bash.

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    File Scopes

    .journal/**

    Shell, read and write are used only for the project's own .journal/ folder: append session entries, rewrite the ≤40-line snapshot, and read back a few lines with head/tail/awk. Nothing outside the project is touched and nothing is sent over the network..journal/**

    Listedtoday

    What's inside

    Frequently Asked Questions

    It never loads the whole history. At the start of work the agent reads at most 40 snapshot lines and the 5 latest index lines (~800 tokens worst case), and recall returns only the 3 newest entries for a tag. The limits are built into the shell commands with head, tail and awk.

    Any agent that can run shell commands in your project: Claude Code, Codex CLI, Cursor, Gemini CLI, Copilot and others. Commands are POSIX shell (bash or zsh); on Windows use Git Bash.

    Starting work costs at most ~800 tokens (40 snapshot lines + 5 index lines). Recording an entry is ~150–300 tokens. Recalling a task returns 3 entries.

    No. There are no scripts or dependencies. The first time you record an entry, the agent creates a .journal/ folder in your project with its own .gitignore, so the journal stays local and is never committed.

    Not in this skill-only version: the agent reads the snapshot when it starts working in a project with a .journal/ folder. For automatic loading at session start in Claude Code, use the free open-source plugin (one SessionStart hook): github.com/IvanStrazov/claude-journal

    New in Productivity

    Free