Dagu Workflow Orchestrator

For least privilege, chart CI explicitly limits its token to read-only, while Helm supports a dedicated service account, RBAC, SSO, and existing Secrets. The product can nevertheless execute local shells, SSH commands, containers, Kubernetes jobs, and MCP operations, with no demonstrated universal runtime sandbox or per-operation permission model, so it scores 2. Human confirmation is supported through human.task, approval, and preview/apply flows, but it is not mandatory for every high-impact command, so it scores 2. Data flows through local files, HTTP/MCP, gRPC workers, third-party actions, external agents, and credential locations are described reasonably well; a complete destination or telemetry inventory is absent, so transparency scores 2. Secret providers, log masking, existing Kubernetes Secrets, and redaction guidance address sensitive data, but implementation, encryption, and default-exposure details are not shown, so this scores 2. Dependencies are versioned, GitHub Actions are commit-pinned, and tools may be hash-pinned; deductions apply for the very large dependency surface, remote-script installation, automatically moving aqua registry, and no shown vulnerability-scanning policy, yielding 2. External effects are explicit in command, SSH, issue-posting, and run-control examples, with optional approvals, but arbitrary commands and MCP execution remain broad, yielding 2. Retries, stopping, history, exit cleanup, and result reuse aid recovery, but there is no general transactional rollback or undo guarantee, so rollback scores 1. The module path, security contact, and a source copyright header provide limited attribution, but publisher identity is unverified and no complete maintainer roster is shown, so source attribution scores 1.

The README, Helm CI, dependency manifest, and health-test hook give broadly consistent accounts of standalone/distributed deployment, authentication, and configuration. Stronger throughput, durable-execution, and reproducibility claims lack benchmarks or implementation evidence, so self-consistency scores 2. Tool pinning, caching, controller model failover, and several installation routes support dependency availability, but execution still depends on registries, external CLIs, cloud SDKs, and a large transitive graph; offline availability is not fully established, so it scores 2. CI emits specific errors and checks expected failures, while workflows expose retries, continue-on-failure, failure handlers, statuses, and logs. Core runtime error implementation and a user-facing error catalog are absent, so failure messages score 2.

The repository identifies its audience and supplies concrete ETL, operations, media, support, edge, container, and agent scenarios, earning 3 for audience and scenarios. It distinguishes standalone, distributed, headless, local-only build workflows, external tools, and licensed features, but does not fully delimit the security boundaries of arbitrary commands, MCP, or model-directed controllers, so capability boundaries score 2. Cron schedules, overlap policies, catch-up windows, dependencies, events, typed parameters, approvals, and explicit start commands make triggering precise, earning 3. Linux, macOS, Windows, Docker, Helm, Kubernetes, SSH, air-gapped environments, and distributed workers are all addressed, with Helm CI covering multiple configurations, so environment fit earns 3.

The README has a clear progression from positioning and architecture to deployment, parameters, and examples, with links to CLI, API, and focused documentation, so information architecture scores 3. Installation covers scripts, Homebrew, npm, Docker, Helm, the service wizard, and RWX storage requirements, earning 3. YAML and action names are mostly consistent and actions support versions, but examples mix id and name and third-party v1 references need not be immutable, so naming stability scores 2. Examples are extensive across common operations; despite no dedicated FAQ, examples and FAQ scores 3. Local-only build execution, environment-dependent capacity, unsupported older releases, and operator responsibilities are disclosed, but there is no systematic limitations or risk inventory, so known limitations scores 2. A full GPL-3.0 text and SPDX source header justify 3 for licensing. Chart CI requires version bumps, the security policy defines supported versions, and dependencies are versioned, but no actual changelog or repository-wide release policy is supplied, so versioning/changelog scores 2. A security email, response targets, disclosure process, and release workflow establish a maintenance path, but publisher identity remains unverified and governance or maintainer details are absent, so maintenance responsibility scores 2.

The Web UI, per-step logs, history, statuses, structured outputs, and artifact flow make results directly inspectable and reusable, earning 3 for output usability. Adding dependencies, retries, approvals, scheduling, and observability around unchanged scripts provides clearly evidenced marginal value over cron, earning 3. A single binary and no required external database reduce deployment cost and multiple deployment options are documented; deductions apply because throughput and operating-cost claims are not execution-verified, distributed operation adds dependencies, and some RBAC/SSO capabilities are in a licensed tier, so cost-benefit scores 2.

Many claims are tied to concrete YAML, architecture descriptions, commands, or configuration names, but performance, security, durability, and reproducibility claims lack corresponding implementation or measurements, so claim traceability scores 2. The README is partially corroborated by Helm CI, SECURITY.md, go.mod, the health-test hook, and the license, but core execution code and the full test suite are not supplied, so cross-source corroboration scores 2. The material distinguishes conditions such as environment-dependent capacity, currently local build workflows, and a moving registry, but several marketing conclusions are not explicitly labeled as unverified, so fact/inference separation scores 2.