Spool now runs Claude Code loops on Anthropic's cloud

Enes Alatas released Spool version 0.6.0 on 26 October at 16:46 UTC under commit 54a1da4, expanding the orchestration hub that runs autonomous Claude Code loops across isolated workstation volumes. According to github.com's report, the update introduces local user authentication, role-based access controls, verifiable binary releases, and tighter process hooks to keep autonomous loops bounded within designated containers. The project coordinates background development loops, tracking execution state alongside token expenditure on Anthropic cloud infrastructure while restricting local file modifications.

This update focuses on platform operators, engineering leads, and multi-tenant development teams deploying persistent coding agents across shared server infrastructure. By replacing open control channels with authenticated sessions, role permissions, and scoped chat bindings, Spool transforms experimental autonomous loops into governed background workers. The following breakdown examines the architectural shifts, security enforcements, and management features introduced in version 0.6.0, as well as the operational telemetry inherited from version 0.5.1.

What shipped in Spool now runs Claude Code loops

Spool version 0.6.0 delivers native identity and access management directly into the orchestration hub, removing unauthenticated administrative control. Operators and engineers sign into the hub using a username and password, while first-time accounts must update a temporary one-time password before accessing workspace assets. Administrative owners manage team accounts through the HTTP API and the browser Settings interface, assigning granular roles that dictate allowed operations. Regular members cannot view owner direct messages, access pending pairing codes, or view conversations outside their assigned scope, preventing privilege escalation across shared development infrastructure.

The core execution environment now provides runtime tracking and sender verification across all active loop sessions. The hub marks every task with a doing trailer, allowing the Fleet overview and individual room timelines to display real-time activity and open system problems. Operators link Slack and Telegram accounts through an explicit confirmation dialog that requires typing an affirmative response before generating an authorized link code. To protect the host environment, an updated spool-hook binary intercepts runtime execution, blocking any Git or GitHub CLI invocations containing embedded secrets and denying process kill commands targeted outside a loop container.

What improved in Spool now runs Claude Code loops

Operational reliability and deployment automation receive substantial upgrades in version 0.6.0, addressing system isolation and secret hygiene across container runtimes. The build pipeline now issues signed binaries accompanied by cryptographic checksums and attestations, while production container images are pinned and published strictly by image digest rather than mutable tags. Internal routing prevents stranger code exhaustion by restricting unauthorized linking attempts to one log entry per hour and pausing compromised surfaces until an administrator intervenes. These enhancements build upon version 0.5.1, which established core budget governors that suspend looping tasks whenever token consumption exceeds preset limits.

System resilience extends across workstation persistence, credential hygiene, and background orchestration scripts. A backup and restore utility covers the local data directory and associated workstation volumes, while administrative runbooks specify precise commands requiring elevated root privileges during state restorations. Upstream toolchain alignment ensures that continuous integration processes enforce pinned golangci-lint installations, daily vulnerability scans, and automated pull request labels on security-sensitive directories. In addition, prompt routing no longer exposes internal private control channels to agent contexts, eliminating accidental leakage of administrative instructions during multi-step code generation.

What you gain from Spool now runs Claude Code loops

Teams deploying Spool gain fine-grained governance over autonomous background workflows without sacrificing the continuous iteration speed of Claude Code. Fleet dashboards present continuous monitoring of token burn across 5-hour and 7-day measurement windows, with individual fleet rows detailing their proportional share of total daily spend. When aggregate activity breaches configured spending ceilings, loops automatically sleep until the rate window resets, protecting organizations against runaway API billing. Operators receive prompt notifications through the hub whenever a background agent encounters persistent failures or when underlying workstation instances become unresponsive.

Security teams benefit from compartmentalized process isolation and defense-in-depth policy enforcement. Because spool-hook denies process kill commands aimed across container boundaries and blocks outbound commands carrying authentication tokens, untrusted code suggestions cannot compromise adjacent tasks or steal repository credentials. In collaborative chat channels, edited messages are ingested as explicit corrections linked to the original reference identifier, keeping loop contexts aligned with human intent. Furthermore, non-interactive onboarding enables headless infrastructure bootstrap, allowing teams to complete system initialization without binding an external chat application.

How to get Spool now runs Claude Code loops

Deploying Spool version 0.6.0 requires provisioning the official release binary or running the preconfigured container image behind a TLS termination proxy as a systemd service. Administrators can fetch verified binaries, inspect cryptographic signatures, and install the underlying Claude Code command-line interface directly onto target host systems. Updating local installations or pulling the latest package releases ensures compatibility with Anthropic model routing and local execution hooks.

npm install -g @anthropic-ai/claude-code
claude update

Once installed, operators configure execution defaults or switch model variants directly within interactive sessions or during process invocation. Selecting specific model targets allows teams to balance reasoning depth against budget limits, while pinning execution flags ensures deterministic behavior across automated loops.

/model claude-fable-5-1
claude --model fable
claude config set --global model claude-fable-5-1

What to watch after Spool now runs Claude Code loops

Following the deployment of version 0.6.0, system operators should monitor credential lifecycle rules, session duration limits, and the administrative Access panel. Session cookies are configured to expire alongside unused background sessions, requiring periodic re-authentication for inactive accounts. Administrators must oversee stranger pairing attempts, as repeated entry of invalid link codes triggers surface pauses that block subsequent account associations until lifted through the management interface. Verifying that the TLS proxy maintains synchronization with dynamic workstation containers remains necessary after Docker daemon restarts.

Engineering workflows must also adapt to the deprecation of direct pairing codes in favor of link codes and explicit multi-step account validation. Because the hub records user modifications across every loop, teams must enforce role hygiene to prevent unprivileged members from altering task assignments. Monitoring CI merge queue ordering and daily security scans will provide early detection of potential container drifts, ensuring that autonomous loops operate reliably without exceeding token quotas or bypassing container boundaries.