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 laurigates/claude-plugins --skill expose-tunable-knobgit clone --depth 1 https://github.com/laurigates/claude-pluginsWrote 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/laurigates/claude-plugins/expose-tunable-knob)<a href="https://agentmods.dev/skills/laurigates/claude-plugins/expose-tunable-knob"><img src="https://agentmods.dev/badge/skills/laurigates/claude-plugins/expose-tunable-knob.svg" alt="Measured on agentmods" 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 Excessive Agency · line 5 Skill selects an external model or provider that may use a different account or billing plan than the operator expects. Undisclosed model switches can cause unexpected cost or quota consumption.Fix: Remove the model/provider override or disclose it prominently and require explicit operator approval before invoking an external coding CLI or billed model.
- medium Excessive Agency · line 45 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
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 | $0.00056 | $0.01377 |
| Opus 5 | $0.00028 | $0.00688 |
| Sonnet 5 | $0.00011 | $0.00275 |
| Haiku 4.5 | $0.00006 | $0.00138 |
Grade A, and why
expose-tunable-knob 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 8d 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 — 96 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Expose Tunable Knob
An agent iterating on a perceptual parameter — a mask's size, a color threshold, an animation's timing, an audio gain — hits a hard wall the moment it can't render, watch, or listen to its own output. A human running a live webcam feed, a rendered UI, or a mixed audio track can judge instantly whether a value looks or sounds right; the agent, reasoning purely from code and possibly one static screenshot, cannot. The naive move — pick a value from domain reasoning, ship it, wait for feedback, repeat — burns a full rebuild/re-run/re-report round-trip per guess, and the agent's guess still carries no more information than the human's own eyes would supply directly.
When to Use This Skill
| Use this skill when... | Skip when... |
|---|---|
| Tuning a value whose correctness is judged by a sense the agent lacks (sight, sound, feel) | The value has an objective, computable correctness criterion (a test asserts the exact number) |
| The user has already pushed back once on a guessed default ("that's better, but...") | This is the first attempt — try a principled default before adding a knob |
| The parameter is genuinely continuous/subjective (position, size, gain, ratio, threshold) | The parameter is binary/structural (a feature flag, an algorithm choice) — that's a decision, not a tuning value |
The runtime already has (or can cheaply gain) a live control surface — a GUI slider, a config‑reload flag, a CLI --watch |
Changing the value requires a full redeploy/recompile cycle with no faster path — a knob doesn't help if it's still one guess per round-trip |
The pattern
- Implement the mechanism, not the magic number. Parameterize whatever currently hardcodes the value — a mask's expansion ratio, a debounce delay, a color-mix weight — so it reads from config/state rather than a literal.
- Pick a reasoned starting default, not an arbitrary one. Use the best available signal: the reference implementation's value, a "just a little more than currently" nudge in the diagnosed direction, or a rough calculation — record why in a doc comment, since the next reader (agent or human) needs the reasoning, not just the number.
- Expose a live control at the layer the human already interacts with —
a UI slider (
egui::Slider, a web form range input), a hot-reloadable config key, a CLI flag re-read per invocation. The requirement is that the human can change it and see the result without asking the agent to redo anything. - Stop guessing values past this point. Once the knob exists, further "should I bump this to 0.3 or 0.4?" turns are wasted — hand the decision to the human and move on to the next piece of work.
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.
- 8d ago First seen · 96 lines · 56 tokens per session scan A a7dcc708db31
expose-tunable-knob is a skill published in the GitHub repository laurigates/claude-plugins (58 stars, last pushed today), licensed MIT. It adds 56 tokens to every session and 1,377 once invoked, about $0.0003 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
design-reference
Skill "design-reference" from animalzinc/claude-plugins, covering design reference, arguments, core rules, process and step 1: receive brief.
html-visual
Generate an interactive single-file HTML visualization — mockup, wireframe, ERD, flowchart, chart, slides, architecture diagram, dashboard, timeline, mindmap, kanban, or table. Use when the user wants something rendered rather than described: a UI mockup or wireframe, a database schema as an ERD, business logic as a…
error-ux
Principles and patterns for writing error messages that help users recover. Use when auditing, writing, or improving error messages in code. Triggers: error messages, user experience, error handling, exception messages, validation errors.
design-compare
Compare Figma designs against implementation screenshots, identifying layout, typography, color, and sizing discrepancies. Generates a structured visual review table and an interactive HTML comparison page with swipe and side-by-side modes. Use when the user asks to compare design with preview, compare Figma with…
sag
ElevenLabs text-to-speech with mac-style say UX.
accessibility
Web accessibility patterns and WCAG 2.1/2.2 compliance for inclusive user interfaces. Use when the user asks to build accessible components, audit a UI for a11y issues, fix screen reader problems, implement keyboard navigation, check color contrast ratios, add ARIA attributes, create accessible forms, or establish…