Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/HenryZ838978/deepseek-harnessnpx agentmods add skills/henryz838978/deepseek-harness/skillWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/henryz838978/deepseek-harness/skill)<a href="https://agentmods.dev/skills/henryz838978/deepseek-harness/skill"><img src="https://agentmods.dev/badge/skills/henryz838978/deepseek-harness/skill/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/henryz838978/deepseek-harness/skill"><img src="https://agentmods.dev/badge/skills/henryz838978/deepseek-harness/skill.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 2 findings, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Data Exfiltration · line 7 Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.Fix: Verify the destination URL is trusted and necessary. Remove or replace with documented APIs. Ensure no secrets, tokens, or PII are transmitted.
- medium MCP Rug Pull · line 139 npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.Fix: Pin the version: npx @scope/[email protected]
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00144 | $0.01958 |
| Opus 5 | $0.00072 | $0.00979 |
| Sonnet 5 | $0.00029 | $0.00392 |
| Haiku 4.5 | $0.00014 | $0.00196 |
Grade A, and why
deepseek-harness scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 160 lines — stays where its author put it; the contents beside it link to each section on GitHub.
DeepSeek V4 Harness
When you call DeepSeek V4-Pro or V4-Flash via the OpenAI-compatible API, you MUST follow the 10 contract rules below. Violating any one of them causes a documented production bug (the upstream issue and our reproduction probe are cited per rule).
The 10 contract rules (ranked by blast radius)
C1 · Disable thinking unless the task is genuinely reasoning-heavy
deepseek-v4-pro defaults to thinking=enabled. Every call then burns 30-300 reasoning_tokens even on trivial prompts.
client.chat.completions.create(
model="deepseek-v4-pro",
messages=messages,
extra_body={"thinking": {"type": "disabled"}}, # ← saves money on every non-reasoning call
)
For TypeScript / openai-node, put thinking at the top level of the request, not inside extra_body (the JS SDK passes unknown top-level keys through):
await openai.chat.completions.create({
model: "deepseek-v4-pro",
messages,
thinking: { type: "disabled" },
} as any);
C2 · In multi-turn loops, preserve reasoning_content on assistant messages
If thinking IS enabled and you re-send a prior assistant message that has tool_calls, you must include the original reasoning_content field. Otherwise the next request returns:
HTTP 400: The reasoning_content in the thinking mode must be passed back to the API.
(Reproduced in reports/probes/probe_2_reasoning_lifecycle.py 3/3 trials on V4-Pro and V4-Flash, 2026-05-09.)
msg = response.choices[0].message
history.append({
"role": "assistant",
"content": msg.content,
"tool_calls": _serialize_tool_calls(msg.tool_calls),
"reasoning_content": getattr(msg, "reasoning_content", None), # ← REQUIRED
})
When a NEW user turn arrives, you MAY strip reasoning_content from prior assistant messages — DeepSeek doesn't require it across user-turn boundaries, and keeping it bloats the prefix-cache key.
C3 · Always set max_tokens (default 4096)
Without an output cap, reasoning_content can stream 8000+ chunks (probes/probe_9_reasoning_runaway.py measured 26 KB / 84 s on a self-doubt prompt) and downstream Electron clients (ChatWise, Cherry Studio) crash with RangeError: Invalid string length once their string buffer hits V8's 512 MB ceiling.
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 160 lines · 144 tokens per session scan A 4343ee57f281
deepseek-harness is a skill published in the GitHub repository HenryZ838978/deepseek-harness (49 stars, last pushed 2d ago), licensed MIT. It adds 144 tokens to every session and 1,958 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
deepseek-harness
Use when building AI agent applications with a plugin-based architecture — Web UI, CLI, Python SDK, Cordis plugin system, multi-model orchestration. DeepSeek Harness (dsh): open-source agent harness by DeepSeek AI where everything is a plugin, powered by Cordis for spatiotemporal composability.
unified-llm-api
Call model APIs through @prismshadow/agenthub — streaming text generation, image generation, speech synthesis, embeddings and the supported-model registry with one client.
llamafactory
Fine-tune LLMs with LlamaFactory — register datasets, train via YAML configs, merge LoRA adapters and serve the result.
vllm
Deploy and serve LLMs with vLLM behind an OpenAI-compatible endpoint, with tool calling enabled for agent workloads.
penguin-sdk
Use whenever the user wants to build an agent application — their own program with an embedded agent, such as an AI app, an agentic app or a RAG app. This is writing application code on the Penguin Harness SDK, not configuring an Agent State inside PenguinHarness. Covers self-contained projects, the createSession/run…
ollama
Deploy and serve local models with Ollama — pull and run them, then expose the OpenAI-compatible endpoint to apps and agents.