Treeship dev

Treeship Dev

Start Here

Use the local zerkerlabs/treeship checkout as the workspace root. Stay scoped to that repo unless the user explicitly asks to inspect sibling repos.

Read these files before code changes:

1. AGENTS.md
2. ONBOARDING.md
3. docs/quality/ai-assisted-development.md

Treat AGENTS.md as the source of truth. If these skill instructions conflict with AGENTS.md, follow AGENTS.md. The code-quality policy in docs/quality/ai-assisted-development.md is not optional; it targets the specific failure modes AI-generated code is prone to in this cryptographic codebase.

What Treeship Is

Treeship is a portable trust layer for AI agent workflows. Actions, approvals, handoffs, receipts, bundles, and related records are signed artifacts that must verify without trusting hosted infrastructure.

The core loop is:

treeship wrap -- your-command
treeship hub push <artifact-id>

Non-Negotiable Invariants

Preserve these exactly:

• DSSE PAE format: "DSSEv1" SP LEN(payloadType) SP payloadType SP LEN(payload) SP payload
• Artifact ID derivation: artifact_id = "art_" + hex(sha256(PAE_bytes)[..16])
• Statement structs do not contain an id field. IDs live on records/sign results.
• Envelope JSON uses DSSE camelCase: payload, payloadType, signatures.
• Approval nonce binding is enforced: action.approvalNonce == approval.nonce.
• Hub/Dock auth uses DPoP with a separate dock keypair. Do not add session tokens.
• Rekor anchoring is best-effort and must not make artifact push fail when Rekor is down.
• TypeScript SDK verification must use the WASM verifier path, not a treeship subprocess.

Repo Map

• packages/core: Rust core library, attestation, statements, keys, storage, verifier, Merkle.
• packages/cli: Rust CLI, command tree, wrap/status/verify/hub/ui/otel commands.
• packages/hub: Go Hub server and SQLite-backed API.
• packages/core-wasm: Browser verifier.
• packages/sdk-ts: TypeScript SDK.
• packages/sdk-python: Python SDK.
• bridges/mcp: MCP bridge.
• docs: Fumadocs documentation.
• skills: portable agent skills.
• plugins: Codex plugin candidates.

Preferred Read Order

For core protocol or CLI work, read:

1. packages/core/src/attestation/pae.rs
2. packages/core/src/attestation/sign.rs
3. packages/core/src/attestation/verify.rs
4. packages/core/src/statements/mod.rs
5. packages/cli/src/main.rs
6. packages/cli/src/commands/wrap.rs

For Hub work, read packages/hub/main.go and the relevant package under packages/hub/internal.

For docs-only work, still read AGENTS.md, then edit only the relevant pages under docs.

CLI UX Rules

When changing CLI behavior:

• Keep commands noninteractive.
• Provide --format json where command output needs stable machine-readable output.
• Return useful exit codes: 0 ok, 1 error, 3 not initialized, 4 usage.
• Include a concrete fix in errors.
• Print a dim next-step hint on success.
• Respect NO_COLOR and --no-color.
• treeship wrap must propagate the subprocess exit code.

Validation

Choose focused checks for the files changed:

cargo test -p treeship-core
cargo test -p treeship-cli
cargo test

For Hub:

cd packages/hub
go test ./...
go build ./...

For TypeScript or MCP packages, inspect package scripts first, then run the relevant npm test/build command from that package directory.

Working Style

Keep edits narrow. Do not refactor unrelated crypto, storage, or schema code while solving a task. Avoid changing generated IDs, signed payload shape, schema field casing, or verification semantics unless the user explicitly asks and the change is reconciled with AGENTS.md.

The full rules for scope, silent fallbacks, verifier preconditions, wire-bound dispatch fields, TOCTOU, RNG, test rigor, and theater commits live in docs/quality/ai-assisted-development.md. Follow that policy.

Use direct, concise writing. Avoid corporate phrasing.