Treeship 0.8: zero to receipt in under 90 seconds
What shipped
Treeship 0.8.0 is about the first ten minutes. treeship add detects the agent frameworks installed on the machine, Claude Code, Cursor, Cline, Hermes, and OpenClaw, and instruments each one. treeship quickstart is a guided interactive setup that goes from nothing to a verified receipt in under 90 seconds. A one-line installer at treeship.dev/setup installs the CLI, initializes a project, and runs treeship add.
The release also gave agents an identity of their own and gave operators a way to watch a session while it runs.
treeship agent registerproduces an Agent Identity Certificate as a.agentpackage with acertificate.html.treeship session status --watchopens a live terminal TUI showing agents, events, security, and verification progress.treeship declarecreates.treeship/declaration.jsonwith the tool authorization scope; receipts now show declared against actual usage and flag unauthorized calls.treeship session eventappends structured events, used by the MCP and A2A bridges.TREESHIP_PROVIDERrecords provider attribution:anthropic,openrouter, orbedrock.- Integration packages for Claude Code, Hermes, and OpenClaw ship under
integrations/, with skill files and MCP configs, plus a universalTREESHIP.mdfor any agent that reads markdown instructions. preview.htmlverifies Merkle structure on its own through Web Crypto, and gained a three-panel narrative, agent cards, timeline grouping, retry detection, approval gates, and a print stylesheet.
Why it matters
Before 0.8, setting up Treeship meant reading docs, editing framework config by hand, and knowing which commands to run in which order. Each framework had its own file to touch. A new user could install the CLI and still have no receipt an hour later.
Now the CLI does the detection. Error messages tell the user which command fixes the problem. treeship init prints the Ship ID, the Key ID, and the next step. treeship wrap without an active session warns and says how to start one. session close opens preview.html in the browser on macOS and Linux, so the first receipt is something the user sees, not a file path.
How it works
treeship quickstart
treeship add
treeship agent register
treeship declare
treeship session status --watch
treeship session closequickstart runs the steps interactively. add looks for each supported framework and writes its integration: skill files and MCP config that point the agent at @treeship/mcp. The MCP bridge now emits session events, so every tool call appears in the receipt timeline, and a failed tool call is audited rather than vanishing from the trail.
agent register signs a certificate for a named agent. The name is sanitized to alphanumeric characters, dashes, and underscores, which closed a path traversal found in review. The .agent package includes certificate.html so the certificate is readable without tooling.
declare writes the authorization scope. When the session closes, the receipt compares the declared scope against the tools the session actually called. A call outside the scope is flagged in the receipt.
session status --watch renders the live TUI. Every event field is sanitized before display, which closed a terminal escape injection, and string truncation is UTF-8 safe. A raw mode guard restores the terminal on every exit path.
On the hub side, the device code now displays all 16 characters and the hub accepts an 8-character prefix for older clients. device_code is redacted from access logs and validated before any database lookup. The hub reads DATABASE_PATH for SQLite persistence, caps session id length, rate limits, and returns consistent JSON errors. Receipts are write-once with atomic first-write ownership and a 10 MB upload limit. The changelog counts more than 15 findings across four rounds of Codex adversarial review, all addressed.
What it does not do
The declaration flags unauthorized tool calls in the receipt. It does not stop them. TREESHIP_COST_USD and the cost_usd field were removed: cost is a consumer concern for dashboards and billing tools, and receipts store verifiable token usage only. preview.html reports "Merkle structure verified", not "Verified", because it checks structure and nothing about the truth of each event. Framework detection covers the five frameworks named above; anything else gets TREESHIP.md and manual wiring.
Where to go next
- Quickstart and First run: the guided path this release created.
treeship add: CLI reference for framework detection.- Agent identity: what an Agent Identity Certificate asserts.
treeship session:status --watch,event, andclose.- Treeship 0.7: session receipts you can verify offline: the package this onboarding leads to.
This entry was written on 11 September 2026 from the 0.8.0 changelog and the code as released, and is filed under the release date.