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 Rheinmir/setup --skill dym-harness-sdk-interventionsgit clone --depth 1 https://github.com/Rheinmir/setupWrote 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/rheinmir/setup/dym-harness-sdk-interventions)<a href="https://agentmods.dev/skills/rheinmir/setup/dym-harness-sdk-interventions"><img src="https://agentmods.dev/badge/skills/rheinmir/setup/dym-harness-sdk-interventions/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/rheinmir/setup/dym-harness-sdk-interventions"><img src="https://agentmods.dev/badge/skills/rheinmir/setup/dym-harness-sdk-interventions.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 | $0.00062 | $0.00719 |
| Opus 5.5 | $0.00025 | $0.00288 |
| Sonnet 5.5 | $0.00012 | $0.00144 |
| Haiku 4.5 | $0.00006 | $0.00072 |
Grade A, and why
dym-harness-sdk-interventions 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.
How it starts
The opening of the file, as written. The whole thing — 41 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Skill: dym-harness-sdk-interventions — chốt chặn tool call
When to use
- Cần agent hỏi người trước khi chạy tool nguy hiểm, hoặc tự phân loại rủi ro, hoặc tuân policy Cedar.
- Khác hook chặn thường: handler trả về QUYẾT ĐỊNH có kiểu (Proceed · Deny · Guide · Interrupt · Transform), registry gộp theo ưu tiên Deny > Interrupt > Transform > Guide > Proceed; Deny cắt ngắn chuỗi, Guide cộng dồn phản hồi, có audit log chung (design
team/designs/0007-intervention-primitive.md).
Steps
- Grammar chuỗi là TẤT ĐỊNH, không đoán nội dung (
harness-py/src/strands_harness/interventions.py):giá trị thành "off"/Nonekhông có handler "ask"HumanInTheLoop(ask=...)— hỏi người mỗi call"smart"HumanInTheLoop(classifier=True)— LLM phân loại rủi ro, chỉ hỏi call bị cờchuỗi kết thúc .cedarCedarAuthorization(policies=<file>)— cầnpip install 'strands-agents[cedar]'chuỗi khác chính sách chữ → prompt của bộ phân loại rủi ro instance InterventionHandlerđi thẳng, không đổi - Ghép nhiều lớp khác loại:
create_harness(interventions=["policies/tools.cedar", "smart"]) - Kiểm hành vi lỗi (đo 2026-09-28, rc 0):
create_harness(session=False, memory=False, interventions=["ask", "ask"]) # ValueError: Two interventions share the handler name 'strands:human-in-the-loop' ... create_harness(session=False, memory=False, interventions=123) # ValueError: Invalid interventions value 123; expected a preset name, ...
Rules
- Hai handler CÙNG loại → raise (SDK chỉ giữ một handler mỗi
name); ghép KHÁC loại (Cedar + một preset người duyệt) thì được. - Policy Cedar viết inline KHÔNG được tự nhận diện (trông như văn xuôi) → truyền
CedarAuthorization(policies=...)trực tiếp. - Subagent con kế thừa interventions của cha; lời gọi tool bên trong
programmatic_tool_callervẫn đi qua interventions — không lách được bằng cách gọi gián tiếp. - Xếp handler rẻ trước (Cedar → dịch vụ → LLM) vì Deny cắt ngắn chuỗi.
"ask"vớiask=Nonedùng stdio — chạy headless/CI sẽ kẹt chờ input; truyền callback hoặc dùng interrupt/resume.
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 · 41 lines · 62 tokens per session scan A 53af11d9a9f6
dym-harness-sdk-interventions is a skill published in the GitHub repository Rheinmir/setup (11 stars, last pushed 20d ago), licensed MIT. It adds 62 tokens to every session and 719 once invoked, about $0.0002 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 2026-11-07.
dym-harness-sdk-interventions compared with skills from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| dym-harness-sdk-interventionsRheinmir/setup | Skill | 11 | 20d ago | MIT | A | 62 | Claude Code |
| imagegenopenai/codex | Skill | 134k | today | Apache-2.0 | A | 113 | Codex |
| local-ai-agentsmicrosoft/ai-agents-for-beginners | Skill | 79k | yesterday | MIT | A | 200 | Claude Code, Codex |
| playwright-climicrosoft/playwright | Skill | 100k | yesterday | Apache-2.0 | A | 19 | Claude Code |
| browser-automationopenclaw/openclaw | Skill | 394k | today | MIT | A | 36 | Claude Code |
| systematic-debuggingobra/superpowers | Skill | 314k | 4d ago | MIT | A | 21 | Claude Code |
| next-cache-components-adoptionvercel/next.js | Skill | 143k | today | MIT | A | 95 | Claude Code |