sandbaseai/deepseek-harness-handbook

Not a software package — browse the live operator site (sandbaseai.github.io/deepseek-harness-handbook) or clone the repo to read the docs.
Contributors run npm run check before submitting (CONTRIBUTING.md).

Usage

• •Route a failure in 30 seconds with the interactive Failure Router, or check the rc.6 Field Status page.
• •Start with the five-minute quickstart, the cheat sheet, or the official examples map.
• •Find the right troubleshooting page by symptom (context window exceeded, ERR_HTTP2_INVALID_SESSION, MCP server not connecting, Windows, PTY shell path, plugin recovery).
• •Use the version-evidence page to capture the exact package/source revision that ran.

Key features

• ✓Agent-first mental model (task boundary, allowed effects, completion condition, model route, tool surface, permission policy, session strategy, failure behavior).
• ✓Source-backed: version-sensitive claims link to official docs or source; pages name verification dates.
• ✓Operational guides with success evidence, failure branches, and safety boundaries; architecture pages prioritize diagrams.
• ✓Multilingual by design — English canonical (28 pages), 简体中文 reviewed (navigation + 3 core guides), ja/ko/es draft.
• ✓Independent project maintained by SandBase — not an official DeepSeek AI project; Apache-2.0.

Best for

Developers and operators building agents on DeepSeek Harness who need a source-backed troubleshooting and architecture reference.

Caveats

• ⚠DeepSeek Harness is in developer preview and may introduce compatibility-breaking changes — pin the revision you deploy.
• ⚠Translation coverage varies; English remains the source of truth until translations point to the current canonical revision.
• ⚠Editorial boundary: no official DeepSeek recommendation, compatibility guarantee, or security endorsement implied for linked resources.