Skip to main content
skill CX Codex

session-logger — Claude Code skill: save session summaries

Writes a short, redacted summary of an agent session: what was decided, which files were touched, what stayed open. Useful when you pick the work up later.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add zhaono1/agent-playbook --skill session-logger -a claude-code
Open on GitHub

What session-logger does

The skill produces a compact recovery artifact instead of treating raw conversation history as durable memory. It fires in two cases: you explicitly ask to record the session, or a supported host sends a session-end event with a transcript path. What lands in the file is bounded — message counts, your last prompt after redaction, up to 12 detected commands and file references, up to 8 questions, a hashed session reference and a project identity.

In Claude the install gives you the agent-playbook session-log command. By default files go to a private local directory at ~/.agent-playbook/sessions/<project-id>/YYYY-MM-DD-<topic>.md. Move the data root with AGENT_PLAYBOOK_DATA_DIR or the --data-dir flag. Writing inside a repository is opt-in: apb session-log --session-dir ./sessions.

That last part deserves a check. If you point the skill at a directory inside a repo, verify its ignore and retention policy first — never assume another project already ignores sessions/. Before writing, the skill strips credentials, bearer tokens, private keys, email addresses and password assignments. Treat that as risk reduction, not a guarantee. Summaries are written with owner-only permissions where the platform supports it and stay local unless you separately authorize a commit or upload.

What it helps with

  • record session outcomes to resume work later
  • hand off context to the next task in a short file
  • capture decisions and open questions
  • list touched files and executed commands
  • strip tokens and passwords before saving

Example prompts for the agent

  • save the summary of this session
  • record what we decided and which files we changed
  • write a handoff summary for the next task

How to install session-logger

1Skills CLI (Vercel)

npx skills add zhaono1/agent-playbook --skill session-logger -a claude-code

2Manually: put the files in place

~/.claude/skills/session-logger/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/session-logger/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about session-logger

How do I install session-logger in Claude?

Run npx skills add zhaono1/agent-playbook --skill session-logger -a claude-code. The skill lands at ~/.claude/skills/session-logger/SKILL.md.

What does session-logger do?

It saves a bounded, redacted session summary: message counts, the last prompt, detected commands and files, open questions. Raw transcripts are not persisted.

Where do session files go?

By default to a private local directory at ~/.agent-playbook/sessions/<project-id>/YYYY-MM-DD-<topic>.md. Repository-local storage is opt-in via --session-dir, and then you should check the ignore policy.

Source text

SKILL.md · MIT
---
name: session-logger
description: Save a bounded, redacted session summary when the user asks to record a conversation or when a supported host provides an explicit session-end transcript event.
allowed-tools: Read, Write
---

# Session Logger

Create a compact recovery artifact without treating raw conversation history as
durable memory.

## Use This Skill When

- The user explicitly asks to save or record the current session.
- A supported host invokes the Agent Playbook `SessionEnd` hook with a transcript path.
- A later task needs a concise handoff artifact and the user has authorized writing it.

Do not silently persist raw transcripts, credentials, private customer data, or
unbounded agent output. Host support and write authority must be visible; skill
metadata alone does not execute a hook.

## CLI Runtime Contract

Claude Code installation can invoke:

```bash
agent-playbook session-log
```

The command reads the host-provided JSONL transcript and stores only bounded,
redacted details:

- message counts;
- the last user prompt, truncated after redaction;
- up to 12 detected commands and file references;
- up to 8 detected questions;
- a hashed session reference and project identity.

By default, files are private local artifacts:

```text
~/.agent-playbook/sessions/<project-id>/YYYY-MM-DD-<topic>.md
```

Set `AGENT_PLAYBOOK_DATA_DIR` or pass `--data-dir` to move the private data root.
Writing inside a repository is opt-in only:

```bash
apb session-log --session-dir ./sessions
```

If a repository-local directory is selected, verify its ignore and retention
policy before writing. Never assume another repository ignores `sessions/`.

## Manual Summary Contract

When the host cannot provide a transcript event, create a user-authorized
summary from the context actually visible in the current session. Include only:

1. outcome and current state;
2. decisions that affect future work;
3. relevant files or commands;
4. open questions and next proof gate.

State omissions honestly. Do not claim full conversation capture, structured
decision extraction, automatic append behavior, or cross-session recall unless
the active runtime demonstrably provides it.

## Privacy and Safety

- Redact common credentials, bearer tokens, private keys, email addresses, and
  password/secret assignments before writing.
- Replace the current project root with a project identity in generated metadata.
- Write generated summaries with owner-only permissions where the platform supports it.
- Keep the artifact local unless the user separately authorizes commit, upload, or sharing.
- Treat redaction as risk reduction, not a guarantee that arbitrary confidential text is safe.

## Completion Check

- [ ] Destination and write authority are clear.
- [ ] The summary is bounded and contains no known raw secret.
- [ ] Repository-local storage was explicitly selected and ignore policy checked.
- [ ] The returned path exists and is readable by the user.

© zhaono1 · MIT · zhaono1/agent-playbook

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set18

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor