setup-agent-logs

name	setup-agent-logs
description	First-run setup for the session log archive. Creates a private Agent-Logs repo on GitHub, clones it locally, and saves the path to memory so handovers auto-archive there. Triggers on phrases like "set up agent logs", "configure handover logging", "where should handovers go", "setup session logs".

Setup Agent Logs

One-time setup that creates the private repo where session handovers are archived, and persists the location so the handover-writer agent can find it in future sessions.

When to use

• First time using the session-transfer plugin on a new machine
• User says "set up agent logs", "configure handover logging", "setup session logs"
• The handover-writer can't find the archive repo and suggests running this

Migration from legacy path (one-time)

Before resolving the data directory, check for legacy data at these paths:

• ~/.claude/projects/-home-*/memory/reference_agent_logs.md

If a legacy file exists AND <plugin-data-dir>/references/agent_logs.md does NOT exist, move it to the new location and delete the legacy file. Tell the user: "Migrated reference_agent_logs.md from ~/.claude/projects/.../memory/ to ."

Path resolution

Resolve the plugin's data directory as $CLAUDE_USER_DATA/dev-tools/ if CLAUDE_USER_DATA is set; otherwise $XDG_DATA_HOME/claude-plugins/dev-tools/ if XDG_DATA_HOME is set; otherwise ~/.local/share/claude-plugins/dev-tools/. Create the directory (and a references/ subdirectory) if it doesn't exist. See the canonical convention in the claude-rudder:plugin-data-storage skill.

Procedure

1. Check if already configured

Check <plugin-data-dir>/references/agent_logs.md for a saved Agent-Logs archive path. If it exists, read it and confirm the path is still valid:

ls <saved-path>/README.md

If the path exists and is a git repo, tell the user it's already set up and show the path. Done.

2. Ask where to store logs

Ask the user where they want the Agent-Logs repo. Suggest a default based on their filesystem:

• If ~/repos/github/my-repos/ exists: suggest ~/repos/github/my-repos/Agent-Logs/
• Otherwise: suggest ~/Agent-Logs/

The user can accept the default or provide a custom path.

3. Check if the repo already exists on GitHub

gh repo view danielrosehill/Agent-Logs --json name 2>/dev/null

If it exists, clone it to the chosen path and skip to step 5.

4. Create the repo

Create the directory, initialize it, and push to GitHub as a private repo:

mkdir -p <path>
cd <path>
git init
git branch -m main

Write the repo by cloning the template from the existing danielrosehill/Agent-Logs repo on GitHub if it exists. If not, scaffold manually with these files:

• CLAUDE.md -- required -- defines the repo's purpose, structure, JSON schema, and how to query. Seed it with:

# Agent-Logs

Private archive of Claude Code session handovers and agent logs.

## Purpose

This repo stores structured JSON session logs produced by the handover-writer agent every time a session handover is created. It serves as a queryable history of all Claude Code sessions -- what was worked on, what's pending, what's blocked.

## Structure

\`\`\`
DDMMYY/
  HHMMSS_model-name.json     Structured session log (one per handover)
sessions.db                   SQLite database (derived, gitignored)
scripts/
  ingest.py                   Scans JSON files → inserts into SQLite
\`\`\`

## How logs arrive

The session-transfer plugin's handover-writer agent automatically writes a JSON file here after every handover. The agent then runs python3 scripts/ingest.py to update the SQLite index.

## JSON schema

Each log file contains: created, model, agent_type, working_directory, repository, branch, last_commit, blocked, user_input_needed, debugging_required, pending_tasks, goal, completed, in_progress, not_started, blockers, resumption_summary.

## Working in this repo

- Don't edit JSON files -- they are append-only records written by agents
- Regenerate the DB at any time: python3 scripts/ingest.py
- Query with SQLite: sqlite3 sessions.db "SELECT created, model, goal FROM sessions ORDER BY created DESC LIMIT 10"
- The DB is gitignored -- it's a derived artifact rebuilt from JSON files

• README.md -- documenting the JSON + SQLite structure
• .gitignore with .DS_Store, .env, sessions.db
• scripts/ingest.py -- the ingestion script that scans DDMMYY/*.json and inserts into sessions.db

The ingestion script, CLAUDE.md, and README are maintained in the canonical danielrosehill/Agent-Logs repo -- prefer cloning over scaffolding from scratch.

Commit and push:

git add -A
git commit -m "Initial scaffold for agent session logs"
gh repo create danielrosehill/Agent-Logs --private --source=. --push

5. Save to memory

Write a reference file so future sessions know where the archive lives:

File: <plugin-data-dir>/references/agent_logs.md

---
name: Agent Logs archive location
description: Path to the private Agent-Logs repo where session handovers are archived by the handover-writer agent
type: reference
---

Agent session logs are archived at `<absolute-path>`.

GitHub repo: `danielrosehill/Agent-Logs` (private).

Structure: `DDMMYY/HHMMSS_model-name.md` folders organized by date.

(The reference is read directly from this path; no global memory index is needed.)

6. Confirm

Tell the user:

• The repo is created at <path>
• GitHub URL (private)
• Handovers will now auto-archive there as DDMMYY/HHMMSS_model.json files
• The ingestion script (scripts/ingest.py) builds a queryable SQLite DB from the JSON logs
• They can run this skill again on another machine to set up the same archive