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 skills/handarbeit/fabrik/write-specnpx skills add handarbeit/fabrik --skill write-specgit clone --depth 1 https://github.com/handarbeit/fabrikWrote 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/handarbeit/fabrik/write-spec)<a href="https://agentmods.dev/skills/handarbeit/fabrik/write-spec"><img src="https://agentmods.dev/badge/skills/handarbeit/fabrik/write-spec.svg" alt="Measured on agentmods" 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 | $0.00157 | $0.06543 |
| Opus 5 | $0.00078 | $0.03272 |
| Sonnet 5 | $0.00031 | $0.01309 |
| Haiku 4.5 | $0.00016 | $0.00654 |
Grade B, and why
write-spec scanned grade B with 1 finding 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 4d 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.
Subtle steeringmediumPrompt injection
Instructions that bias recommendations or shape behaviour without the user noticing.
- **What happens next**, described as something they can say rather than something they must run: "if you want, I can go back over this and tighten anything vague — just ask", or "this is ready to hand to whoever's build How it starts
The opening of the file, as written. The whole thing — 324 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Create a Feature Specification
Produce a feature specification a development team can build from, using GitHub Spec Kit's structure and quality bar. The specification describes what users need and why it matters — never how to implement it.
The person using this is a product manager, business owner, or domain expert — deep knowledge of the business, none of software construction assumed. Write the spec for business stakeholders and keep the conversation in plain language. Never ask them a question only an engineer could answer, and don't use a technical term without defining it in the same sentence. Don't narrate file paths or internal steps as you go — but do say where the spec ended up when you're finished, so they can find it.
How to ask anything, anywhere in this skill
This governs every question below, not just the clarifying questions in step 6:
- Always lead with a recommendation. A neutral menu hands the decision back to someone who came here precisely because they don't know the options. Say what you'd do and why, in a sentence, and make agreeing cheap: "say yes and I'll take it."
- "You pick" / "I don't know" / "which is normal?" is an answer. Take your recommendation and carry on — never stall waiting for a preference they don't have. Record the result under Assumptions as your guess, not as their decision.
- Never invent a specific number, threshold, name, or percentage they haven't given you or clearly implied. Filling in structure is the job; manufacturing facts is not. See "Writing success criteria" for what to do when a figure is genuinely missing.
- Batch what you can. Every separate round-trip costs the person patience they'd rather spend on the spec.
Where the work goes
Specs live in a folder the engineering team can read. Before writing anything, establish the working folder:
- If exactly one folder is connected to this session, use it.
- If several are connected, ask which one to use — don't say "which one holds the specs", since on a first run none of them does.
- If no folder is connected, say so plainly — "I don't have access to a folder I can save this in, so I'll hand these back to you to download" — and hand the files back instead. Everything below still applies: produce all three deliverables, at these relative paths, so they can be dropped into a project intact:
specs/NNN-short-name/spec.mdspecs/NNN-short-name/checklists/requirements.md.specify/feature.json— note this is a sibling ofspecs/, not inside the feature folder, so handing back only the feature folder leaves it out
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.
- 4d ago First seen · 324 lines · 157 tokens per session scan B 919b3f1e148d
write-spec is a skill published in the GitHub repository handarbeit/fabrik (22 stars, last pushed today), licensed Apache-2.0. It adds 157 tokens to every session and 6,543 once invoked, about $0.0008 per session on Opus 5. A static security scan graded it B with 1 finding (subtle steering). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
frame-macos-notification
拟真 macOS 通知 banner + app icon + 标题正文, 适合 video overlay / 产品发布预告.
deck-course-module
暖纸背景 + Playfair, 左侧学习目标常驻, 含 MCQ 自测页.
deck-pitch
10 页融资 deck, 白底 + 蓝紫渐变 hero, traction 柱状, $X.XM ask.
argent-tv-interact
Control and inspect TV apps via argent — Apple TV (tvOS), Android TV (leanback), and Amazon Fire TV (Vega). Boot the target, read focus, navigate with the D-pad remote, type, screenshot, and on Vega debug the JS runtime (evaluate, console logs, network inspector). Use when a task targets a TV (runtimeKind "tv", or…
review-offered-task
Review a task that has been offered to you and decide whether to accept or reject it.
aidd-dev:08:for-sure
Iterative agent loop that tracks attempts and retries until a success condition is met. Use when the user says "for sure", "make sure", "keep trying until", "loop until done", "don't stop until", or needs guaranteed completion of a task with explicit success criteria.