DeepSeek Bot

The README and role design distinguish the management scope of the user, Main Bots, group admins, and ordinary Bots. Users set sandbox presets, full permissions require another confirmation, and local MCP commands require approval. The Docker note says Bots can access mounted content, while the permissions plan records a DSH menu path that can bypass Bot presets, so least privilege and confirmation are not comprehensive. Data documentation covers local storage, group memory boundaries, and data retention after uninstall, but the supplied material shows no dedicated sensitive-data protection or encryption. The release workflow uses OIDC provenance; the test workflow explicitly disables npm audit during npm ci, and the evidence does not describe dependency vulnerability handling. Failed archives retain sessions for retry; acknowledgments and the independent-project disclaimer clearly identify inspirations and attribution. Publisher identity is unverified and treated as unknown, with no score adjustment.

The workflows specify checks across operating systems, Node and DSH versions, packaging, and Docker; supplied tests cover Agent sessions, permission restrictions, archive failures, and error returns. README, the permissions plan, and tests are broadly consistent, but this static evidence cannot establish that checks passed. The project depends on DSH, Node/pnpm or the npm registry, with some alternatives such as offline packages or a mirror; it remains subject to external dependency availability. Tests show specific errors for some operations and preserve state on failure. Not all failure paths have documented user messages, so the scores are not higher.

The README describes individual Bots, a main Bot, group chats, memory, a capability library, and usage views. The permissions plan specifies roles, sandbox presets, and some explicit limitations. Group responses default to an admin and users can address Bots with @, giving reasonably clear triggers, though not every trigger condition or boundary is documented. Installation covers Windows, macOS, Linux, and Docker/NAS, with separate instructions for the desktop app, DSH plugin, and web environment.

The README is organized around installation, setup, features, data, and development, and links plugin documentation, contribution guidance, the changelog, and Docker instructions. Installation steps and platform requirements are concrete. Product and directory names are generally clear, though DSH and DS Bot version numbers differ in some places and readers must follow the versions in each command. Screenshots and procedural examples are present, but there is no systematic FAQ. The permissions plan records a known DSH menu bypass; the README advises checking important actions, while other limitations are sparsely documented. The Apache-2.0 license is included, and version badges, release checks, and a changelog provide version context. CI and contribution guidance show a maintenance process, but a responsible maintainer or long-term support commitment is not explicit. Unknown publisher identity is not itself a deduction.

Bots, group handoffs, group memory, and the capability library support team workflows, and the README describes usage tracking. It states that group chats cost more, Bots use the user's model quota, and can read or write files or run commands when asked. It does not quantify costs, quality gains, or task boundaries, so these scores remain moderate.

Core features and boundaries can be traced to the README, permissions plan, workflows, and specific tests; the tests provide complementary evidence for Agent creation, archive failures, permission masks, and error responses. Sources corroborate some claims, but the supplied material omits complete implementation sources and cannot establish CI results or runtime behavior. Documentation distinguishes stated behavior from some known limitations, but some capability and safety claims remain project assertions without independent validation, so full marks are not justified.