Session Kit

A composable set of Claude Code skills for managing session lifecycle — from starting work, through the session, to parking it and finding it later.

Why

Claude Code sessions are ephemeral. When you close a session, the context is gone. Session Kit gives you a lightweight system for:

• Priming repos with expert skills and context files for productive sessions
• Parking sessions with structured artifacts (summary, resume context, prompt analysis)
• Resuming sessions with zero re-explanation
• Finding past work across projects via a searchable index
• Sharing results with teammates in multiple formats (quick summary, full write-up, investigation package)

Install

git clone https://github.com/jstoobz/session-kit.git
cd session-kit
./link.sh

This symlinks each skill into ~/.claude/skills/ and the sk dispatcher binary into ~/.local/bin/sk. Restart Claude Code to pick up the new skills.

The sk binary is the backing implementation for every migrated skill (/checkin, /index, /park, /tldr, /relay, /hone, /retro, /handoff, /rca, /persist). Each SKILL.md is a thin orchestrator that invokes a sk <subcommand>. Run sk --help for the full subcommand reference. ~/.local/bin must be on your PATH.

Requires Python 3.11+ and uv. The bin/sk shebang invokes uv run to manage its own dependencies (typer, filelock) — no system-wide install needed.

Skills

Registration — register the session before any artifact is written

Command	Purpose
/checkin	Register the current Claude Code session in manifest.json, pre-allocate the durable archive dir + ledger. Invoked silently as a precondition by every artifact-producing skill below.

Core Artifacts — generate session documents

Command	Output	Purpose
/tldr	.stoobz/TLDR.md	Concise session summary — key findings, decisions, changes
/relay	.stoobz/CONTEXT_FOR_NEXT_SESSION.md	Everything needed to resume in a new session
/hone	.stoobz/HONE.md	Original prompt + analysis + optimized version
/retro	.stoobz/RETRO.md	Session retrospective — what went well, what to improve
/handoff	.stoobz/HANDOFF.md	Teammate-facing write-up with full business context
/rca	.stoobz/INVESTIGATION_SUMMARY.md + .stoobz/evidence/	Root cause analysis package for engineer + Claude consumption

Lifecycle — manage session flow

Command	Purpose
/park	End session — generate artifacts, archive, update manifest
/park <label>	Park with explicit archive label (e.g., /park PROJ-1234)
/pickup	Start session — load prior context, present briefing, inherit chain metadata from the relay baton
/persist <name> <tags>	Save a reference artifact mid-session to .stoobz/. Tags merge-dedupe into the manifest entry, surfaced by /index <tag>
/checkpoint	Synthesize selected chain nodes into a new starting point (branch chain)
/index	Find past sessions from manifest
/index --active	Only in-flight (unparked) sessions, with resume commands
/index --orphans	Filesystem active dirs that never registered (legacy recovery)
/index --deep <term>	Search inside archived artifact content

Project Setup — create permanent expert knowledge

Command	Purpose
/prime	Analyze repo, create expert skills + context files
/prime --refresh	Update stale skills based on what changed

Maintenance

Command	Purpose
/sweep	Interactive cleanup of old Claude Code sessions from the resume picker
/park --archive-system	Retroactive cleanup of scattered .stoobz/ directories (pending re-migration; see park/SKILL.md)

Archive Structure

Session Kit archives to a central location (default ~/.stoobz/):

~/.stoobz/
├── manifest.json                          ← searchable index for /index
└── sessions/
    ├── my-project/
    │   ├── <session-id>-active/           ← /checkin pre-allocates this for live sessions
    │   │   ├── .session-artifacts.json    ← ledger of artifact writes
    │   │   └── (in-flight artifacts)
    │   ├── 2026-02-13-PROJ-1234/          ← /park renames -active/ to date-label form
    │   │   ├── .session-artifacts.json
    │   │   ├── TLDR.md
    │   │   ├── HONE.md
    │   │   └── RETRO.md
    │   └── auth-flow-notes.md             ← /persist reference artifact
    └── another-project/
        └── 2026-01-28-rate-limiting/
            ├── TLDR.md
            └── INVESTIGATION_SUMMARY.md

During a session, every artifact is written directly to the durable <session-id>-active/ directory (via sk write-artifact) and copied to ./.stoobz/ for in-session discovery. /park runs sk park-finalize, which renames the active directory to its final <date>-<label>/ form, flips the manifest entry from active to archived, and leaves .stoobz/CONTEXT_FOR_NEXT_SESSION.md as the relay baton for /pickup.

This is the durable-first protocol — if the session terminates abnormally, the artifacts are already in the central archive. See write-artifact-protocol.md for the contract artifact-emitting skills follow.

Configuration

Variable	Default	Purpose
SESSION_KIT_ROOT	~/.stoobz	Root directory for archives and manifest

Set SESSION_KIT_ROOT to change where archives are stored:

export SESSION_KIT_ROOT="$HOME/.sessions"

Quick Reference

I want to...	Use
Register this session before doing anything	/checkin (also auto-fires under every artifact skill)
Set up expert skills for a new repo	/prime
Update stale expert skills	/prime --refresh
Save everything before stepping away	/park
Resume where I left off	/pickup
Branch a new chain from selected past sessions	/checkpoint
Share a quick summary	/tldr
Write up findings for the team	/handoff
Save context for next session	/relay
Improve my prompting	/hone
Reflect on my process	/retro
Package an investigation	/rca
Find a past session	/index
Save a reference mid-session	/persist
Find a session by tag	/index <tag>
List in-flight sessions	/index --active
Surface unregistered active dirs	/index --orphans
Search inside archived content	/index --deep <term>
Clean up old sessions	/sweep

See guide.md for detailed workflows and composability patterns. The sk CLI also exposes machine-readable surfaces (sk checkin --json, sk write-artifact, sk park-finalize --json) for scripting or testing — run sk --help for the full subcommand list.