DeepSeek Harness is an open-source framework for building and running AI agents, organized so that its functionality is provided through plugins. Developers use it to launch agent workflows through a local web interface or from source, while the catalogue add-ons extend the harness as plugins.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add deepseek-ai/deepseek-harness --skill diagnose-windows-sandbox-aclgit clone --depth 1 https://github.com/deepseek-ai/deepseek-harnessWrote 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/deepseek-ai/deepseek-harness/diagnose-windows-sandbox-acl)<a href="https://agentmods.dev/skills/deepseek-ai/deepseek-harness/diagnose-windows-sandbox-acl"><img src="https://agentmods.dev/badge/skills/deepseek-ai/deepseek-harness/diagnose-windows-sandbox-acl/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/deepseek-ai/deepseek-harness/diagnose-windows-sandbox-acl"><img src="https://agentmods.dev/badge/skills/deepseek-ai/deepseek-harness/diagnose-windows-sandbox-acl.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00067 | $0.01534 |
| Opus 5.5 | $0.00027 | $0.00614 |
| Sonnet 5.5 | $0.00013 | $0.00307 |
| Haiku 4.5 | $0.00007 | $0.00153 |
Grade A, and why
diagnose-windows-sandbox-acl 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 16d 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 — 64 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Diagnose Windows sandbox ACL failures
Write every approval request in plain words, in the user's language. The prompt is all the user reads before widening access, so it must stand alone: which folder the script touches, that it adds the signed-in user's full-control entry where a right is missing and removes foreign package entries, that file contents and owners are unchanged, and that the printed recovery command undoes each change. Keep error codes, WRITE_DAC/WRITE_OWNER, S-1-15-2-* SIDs, icacls, ACL, ACE, verdict names and switches out of the request: write Windows file permissions, not ACL or ACE. Ask once, for one command.
One command diagnoses and repairs
The script has no modes. -Path, -AllowRoot and -Out read the path and every ancestor and repair what the observations prove, in the same run:
- directories on that chain that lack effective
WRITE_DACorWRITE_OWNERreceive a full-control allow ACE for the signed-in user, because DSH cannot provision its workspace grant without them; - explicit AppContainer package allow ACEs (
S-1-15-2-*, except the well-known groups ending in 1 or 2) are removed at their sources, ancestor first, which also removes those packages' access; - a directory the sandbox cannot provision — the state its provisioning error reports on the workspace root — or the authorized root itself also has its subtree searched for those ACEs in the same run, so a deeper entry needs no second request. That bounded walk reports
truncatedand unreadable directories; if truncated, pass the still-failing deeper path once more.
Every change is backed up, then verified by re-reading it. -AllowRoot bounds all of it: an object is changed only when it is that directory or strictly inside it, so a workspace root can repair itself. Never split this into a diagnostic call and a repair call, never ask twice for one repair, and never pass a mode switch that does not exist.
& '<skill-directory>\scripts\diagnose-windows-sandbox-acl.ps1' -Path '<failing-path>' -AllowRoot '<authorized-directory>' -Out '<recovery-directory>'
exit $LASTEXITCODE
What ships with it
1 file 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.
- 16d ago Changed · +17 lines · -4 tokens per session 9c616c78892d
- 17d ago First seen · 47 lines · 71 tokens per session scan A 92a6bdc6a47c
diagnose-windows-sandbox-acl is a skill published in the GitHub repository deepseek-ai/deepseek-harness (250,433 stars, last pushed today), licensed MIT. It adds 67 tokens to every session and 1,534 once invoked, about $0.0003 per session on Opus 5.5. 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 2026-09-29.
Other skills, from other repositories
diagnosing-bugs
Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.
manage-taskboard
Manage work in the native DeepSeek Harness Taskboard with exact task ids and optimistic versions. Use when an Agent must inspect project work, claim an eligible todo, record progress or blockers, verify an implementation, submit it for human review, or release its own claim; also use when a human asks how to accept…
industry-research-method
A Chinese-language method for researching an industry by mapping its supply chain, tracking policies and news, and writing sourced reports. It separates the chain into upstream inputs, midstream production, and downstream distribution or use.
company-research-method
A documented process for researching a company using supplied files and selected public sources. It produces a company overview with business details, sourced financial figures, and stated risks.
dsh-troubleshoot
A troubleshooting guide for DeepSeek Harness, including its process, configuration, plugins, tools, skills, and model connections.
ask-good-questions
Guidance for asking technical questions and reporting bugs so another developer can reproduce and answer them. It covers versions, environment details, minimal reproduction steps, expected and actual results, errors, and attempted fixes.