loomlog

loomlog

Local, cross-agent dev journal for Claude Code, Codex, and Gemini CLI. Passively weaves the sessions your AI tools already log into an Obsidian-compatible vault, then lets you recall any day ("what did I do?") and reflect with research-backed frameworks. No API key. Token-cheap.

loom = the three agents' threads, woven into one log + knowledge graph.

Status: published on npm. Claude Code and Codex are fully supported; Gemini CLI is best-effort/experimental (it logs prompts only and auto-deletes sessions). See grill-loomlog-20260607.md for the full locked design and RELEASING.md for the release pipeline.

Why

Your AI coding sessions already get logged to disk — but they scatter across three tools and (for Claude Code & Gemini) get auto-deleted after a while. loomlog captures them promptly into durable Markdown you own, and lets you ask "what did I do today?" across every agent at once.

• Capture is mechanical (no LLM) → 0 tokens. It just parses logs the agents already write. Secrets are redacted before anything is stored.
• Reports run inside your agent via a slash command — the host model formats them, so there's no API key and no extra cost.
• Storage is plain Markdown in a folder. Point it at an Obsidian vault and the graph (Daily ↔ Project ↔ Topic) lights up automatically.

How it works

agent session logs ──(capture, mechanical, 0 tokens)──▶ <vault>/Daily/*.md , Projects/*.md
                                                              │
                                   /report (slash command) ───┘──▶ host model writes your report

Agent	Auto-deletes logs?	Capture strategy
Claude Code	Yes (30d default)	Stop hook → capture immediately
Codex	No	lazy scan at report time
Gemini CLI	Yes (on by default)	scheduled daily scan (experimental)

Recall — "what did I do?"

Forgot what you worked on? Ask loomlog. The first arg is a query (mechanical, 0 tokens, works in a plain terminal):

loomlog 2026-06-08        # a specific day        loomlog today | yesterday
loomlog week              # last 7 days           loomlog month        # last 30
loomlog <project>         # one project's history loomlog patterns     # what work you do most

patterns answers "what kind of work do I do?" — your command-type mix, time split across projects, agent usage, busiest days, and recent commits (loomlog reads your git commit messages straight from the logs — your own "what I shipped" log, 0 tokens).

Reflect — structured retrospection, grounded in the research

Recall is what; reflection is so what / now what. Reflection runs inside your AI agent (Claude Code / Codex / Gemini) as a slash command — because the back-and-forth needs a model, and loomlog's rule is the host model does it, no API key. loomlog mechanically fills the factual layer; the agent walks you through a real reflective-practice framework; the result is saved to Reflections/<date>.md (a folder capture never overwrites).

How a daily reflection actually goes — in Claude Code, after a day of work, you type:

/loomlog:reflect

and the agent:

1. pulls today's facts (loomlog reflect --json under the hood),
2. shows you What you did — per project: intent, key files, work type, and the commits you shipped,
3. asks you the So What questions one at a time — "今日いちばん重要だった作業は？", "詰まったのはなぜ？", "新しく分かった/決めたことは？" — you answer in the chat,
4. asks the Now What — "次にやること/変えることは？",
5. writes the finished reflection to ~/loomlog/Reflections/2026-06-09.md and links it back to that day's note (so it shows up in your Obsidian graph).

/loomlog:reflect          # daily   — What / So What / Now What (Borton→Driscoll)
/loomlog:weekly           # weekly  — Gibbs Reflective Cycle (1988)

Pick a different framework with an argument: /loomlog:reflect aar (After-Action Review, good when the day was blocker-heavy), kpt, or ywt. Under the hood it's just loomlog reflect --template <t> --json → you answer → loomlog reflect-save (≈0 extra tokens, no API key). In a bare terminal without an agent, use the Recall commands above instead — loomlog reflect alone only prints the model-facing JSON.

Quick try (dev)

npm install
# Capture one Claude Code transcript into a vault:
npx tsx src/cli.ts capture ~/.claude/projects/<proj>/<session>.jsonl --vault ./my-vault
cat ./my-vault/Daily/*.md

Install

npm install -g loomlog
loomlog init                 # creates ~/loomlog, writes the Obsidian graph config,
                             # registers the vault with Obsidian, detects your agents
export LOOMLOG_VAULT=~/loomlog