treeship

name	treeship
description	Signed session receipts for every OpenClaw task (v0.9.4)
version	0.9.4
author	Zerker Labs
homepage	https://treeship.dev
openclaw	{"category":"security","tags":["attestation","receipts","verification","trust","mcp"]}
requires	{"bins":["treeship"],"env":["TREESHIP_ACTOR"]}

Treeship

Generate signed, verifiable session receipts for every OpenClaw task. Each receipt captures what happened, who did it, what changed, and provides cryptographic proof that the record is trustworthy.

Current version: 0.9.4 — Rust core, 161 tests passing, Ed25519 DSSE envelopes, offline verification.

Quick Setup

One-liner (macOS/Linux):

curl -fsSL treeship.dev/setup | sh

Or via npm (inspectable, signed package):

npm install -g treeship
treeship init

The setup script auto-detects OpenClaw and installs this skill to ~/.openclaw/skills/treeship/.

Primary workflow

At the start of every task

treeship session start --name "<task description>" --actor agent://openclaw

Optional: set model metadata for richer receipts:

export TREESHIP_ACTOR="agent://openclaw"
export TREESHIP_MODEL="<your model>"

During the task

Every shell command MUST be prefixed with treeship wrap --:

treeship wrap -- npm test
treeship wrap -- cargo build --release
treeship wrap -- python3 script.py
treeship wrap -- curl -sS https://api.example.com/endpoint

For multi-step shell commands:

treeship wrap -- bash -c "git add -A && git commit -m 'fix: resolve bug'"

Sub-tasks

For sub-tasks within a larger task, continue using the same session. Do NOT start a new session. The receipt should capture the entire task as one unit.

When the task is complete

Close with a meaningful summary:

treeship session close \
  --headline "<one-line accomplishment>" \
  --summary "<detailed description of what changed>" \
  --review "<what a reviewer should check before trusting this work>"

Upload for a shareable URL:

treeship session report

Writing good summaries

Good headline: "Fixed JWT expiry bug that caused 401s after 24 hours" Bad headline: "Fixed a bug"

Good summary: "Found the TTL was set to 24h instead of 7d in token.ts. Fixed the constant, added a regression test, and confirmed the fix in local tests." Bad summary: "Fixed the bug."

Good review: "Verify the TTL change works with existing tokens in production." Bad review: "Please review."

What gets captured

Data	How	Status
Shell commands	treeship wrap --	AUTO
Exit codes	treeship wrap --	AUTO
Elapsed time	treeship wrap --	AUTO
File writes	File snapshot diff	AUTO
Model name	TREESHIP_MODEL env var	EXPLICIT
Actor URI	TREESHIP_ACTOR env var	EXPLICIT
Session narrative	--headline/--summary/--review	EXPLICIT
Merkle proof	Automatic	AUTO
Ed25519 signatures	Automatic	AUTO

What does NOT get captured

• Commands run without treeship wrap --
• Raw argument values (only SHA-256 digests)
• Raw output content (only SHA-256 digests)
• File reads (run treeship daemon start for atime detection)
• Internal agent reasoning
• Environment variable values

MCP Bridge (automatic tool-call capture)

If @treeship/mcp is configured as your MCP server, every callTool invocation is automatically attested:

• Intent attestation (signed before the call): actor, tool name, args digest
• Result receipt (signed after): exit code, elapsed time, output digest
• Session event: timeline entry linking intent to result

No treeship wrap -- needed for MCP tool calls. Configure via:

npx -y @treeship/mcp@latest setup

Additional commands

treeship status              # show keys, recent artifacts, hub status
treeship verify last         # verify the most recent artifact
treeship verify <id>         # verify a specific artifact
treeship package verify <path>  # verify a .treeship package offline
treeship hub attach          # connect to treeship.dev Hub
treeship hub push <id>       # push artifact to Hub
treeship ui                  # interactive TUI dashboard
treeship doctor              # diagnostic check
treeship add                 # instrument AI agents in current project
treeship quickstart          # guided first-time setup

Trust model

• Local-first: Everything works offline. Hub is optional.
• Self-contained: A receipt is a JSON file. Verifies without trusting Treeship infrastructure.
• Privacy-aware: SHA-256 digests of args/outputs, not raw content.
• Deterministic: Same content → same artifact ID always.
• Open source: Verifier is open source at github.com/zerkerlabs/treeship

Verification

Verify any receipt locally (pure WASM, no network):

treeship verify art_<id>
treeship package verify .treeship/sessions/ssn_*.treeship

Or via web: https://treeship.dev/verify/

Resources

• Docs: https://docs.treeship.dev
• Source: https://github.com/zerkerlabs/treeship
• CLI help: treeship --help, treeship <command> --help
• Trust inventory: https://github.com/zerkerlabs/treeship/blob/main/TREESHIP.md