Guides
Step-by-step DeepSeek Harness guides: install dsh, boot profiles, configure agent modes, and wire up sub-agents.
DeepSeek Harness agent presets: the four modes are directories
A preset is a directory holding one agent.cordis.yml. The shipped roster, why authoring is copy-only, the produced-nothing switching rule, how a preset's rows resolve, and why a broken preset is listed rather than skipped.
Call Claude Code and Codex as sub-agents from DeepSeek Harness
dsh ships six sub-agent providers, including real Claude Code (via the official Agent SDK) and real Codex (via codex app-server). The exact config rows, why the tools ship disabled, credential scrubbing, and what the child does and does not inherit.
Port a Claude Code skill to DeepSeek Harness (it already works)
dsh reads the same SKILL.md format Claude Code does. Copy the directory into ~/.agents/skills and it is discovered — verified against a running instance. The frontmatter contract, the five discovery roots, and how to confirm it loaded.
The four DeepSeek Harness agent modes, and when each one wins
Standard, Code, Minimal and Creator mode explained — what each exposes to the model, why the four modes are really four agent presets, and how to pick between them.
dsh CLI reference: every subcommand, flag and entry mode
The complete dsh launcher reference — --profile, --port, --patch, --dump-config, --dump-default-config, the plugin subcommand, headless job invocation, $DSH_HOME paths, argument order and exit codes.
Every tool in DeepSeek Harness, and the package that provides it
The complete model-facing tool catalog for dsh — bash, run_code, str_replace_editor, glob, grep, lsp, subagent, workflow, ralph, goals, session queries and the cordis runtime tools — mapped to the packages they ship in.
DeepSeek Harness glossary: turn, step, round, seam, scope, goal
The vocabulary dsh uses and what each term means precisely — the turn/step/round loop hierarchy, capability seams, agent scopes and shadowing, goals and goal rounds, human commands, and the Ralph loop.
Running DeepSeek Harness headless (and in CI)
The headless profile runs one session, emits a result and exits — no server. How to invoke it, why exit codes matter, and how to keep a CI profile separate from your interactive one.
How to install DeepSeek Harness (dsh) — the complete guide
Install the dsh CLI globally or run it with npx, boot the web profile on port 3080, add a model provider, and verify the install with --dump-config. Includes the from-source path.
DeepSeek Harness profiles and patch layers explained
How a dsh profile composes bundles into a running agent: the four-step patch order, cordis.patch.yml at profile and home level, the --patch overlay, and using --dump-config to see what actually booted.
Sub-agents, workflows and the Ralph loop in DeepSeek Harness
How dsh delegates work: the subagent and subagent_fork tools, background agent control, the workflow tool's agent/pipeline/parallel primitives, job management, and what a Ralph loop actually is.