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 agentmods add commands/jhamidun/screencast-desktop/setupgit clone --depth 1 https://github.com/JHamidun/screencast-desktopWhat 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 | $0.00026 | $0.00440 |
| Opus 5 | $0.00013 | $0.00220 |
| Sonnet 5 | $0.00005 | $0.00088 |
| Haiku 4.5 | $0.00003 | $0.00044 |
Grade A, and why
setup scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
Prepare this machine for the screencast-desktop plugin.
- Run the diagnostics and show me its output in full:
python "${CLAUDE_PLUGIN_ROOT}/server/doctor.py"
It checks the things that break silently: display scaling, monitor layout (non-primary monitors have negative coordinates), presence of encoders, whether window capture actually works, and whether a voice for narration is available.
- If it reported missing packages — install them:
pip install -r "${CLAUDE_PLUGIN_ROOT}/requirements.txt"
-
If ffmpeg is not found — you need the full build (the stripped-down ones lack the required filters):
winget install Gyan.FFmpeg, then verify withffmpeg -version. -
Voiceover needs
ELEVENLABS_API_KEY— in the environment variables or in a.envfile in the plugin root (ELEVENLABS_API_KEY=..., optionallyELEVENLABS_VOICE_ID=...). Without it everything else works, the video will just have no voice — say so, do not treat it as an error. -
If the UI automation binary is missing from the plugin — download it. It is deliberately not shipped with the package: it is a third-party project (sbroenne/mcp-windows, MIT), weighs about 60 MB, and is updated separately. The script verifies the checksum against the one published in the release and installs nothing without a match:
python "${CLAUDE_PLUGIN_ROOT}/server/fetch_ui_binary.py"
- Check that both plugin servers are connected:
claude mcp list
screencast and windows-ui should respond.
At the end, say in one line whether the machine is ready to record or what is missing. Do not report success until you have seen the command output.
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.
- yesterday First seen · 53 lines · 26 tokens per session scan A d46b4838734d
setup is a command published in the GitHub repository JHamidun/screencast-desktop (0 stars, last pushed 7d ago), licensed MIT. It adds 26 tokens to every session and 440 once invoked, about $0.0001 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 2026-08-31.
Other commands, from other repositories
run
显式启动一次隔离的 Windows Computer Use 任务.
doctor
检查 Computer Use 插件、环境和 Reasonix 能力映射.
feature
End-to-end feature/bug-sweep workflow for ui-debugger-mcp — understand, reproduce against a real target, explore in parallel, split into path-disjoint slices, build with a hive of agents in this ONE checkout (never worktrees), gate green, PR, merge, release to npm. Tracks in GitHub issues. Reads intent from the prompt.
trace
查看或导出脱敏 Computer Use 轨迹.
planx
Write a concise, self-contained execution plan to docs/plans/ / / / - / for another AI to implement.
benchmark
生成 Computer Use 能力评分报告.