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 buildinternet/releases --skill classify-media-relevancegit clone --depth 1 https://github.com/buildinternet/releasesWrote 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/buildinternet/releases/classify-media-relevance)<a href="https://agentmods.dev/skills/buildinternet/releases/classify-media-relevance"><img src="https://agentmods.dev/badge/skills/buildinternet/releases/classify-media-relevance/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/buildinternet/releases/classify-media-relevance"><img src="https://agentmods.dev/badge/skills/buildinternet/releases/classify-media-relevance.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 | $0.00057 | $0.01309 |
| Opus 5 | $0.00028 | $0.00655 |
| Sonnet 5 | $0.00011 | $0.00262 |
| Haiku 4.5 | $0.00006 | $0.00131 |
Grade A, and why
classify-media-relevance 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 9d 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 — 91 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Classifying Media Relevance
Release pages contain two kinds of media: editorial content that belongs in the release (screenshots of the feature, demo videos, diagrams explaining a change) and site chrome that doesn't (author avatars, nav logos, tracking pixels, decorative separators). This skill governs which items end up in a release's media[] array.
The goal is precision-over-recall: a dropped editorial image is recoverable (users click through to the source page), but a kept junk image pollutes the UI and wastes storage.
When this runs
- During the parse pipeline, after the AI extracts release content from a fetched page.
- During crawl-mode fetches, when the extractor reads full-page markdown from a linked article and produces a fresh
media[]. - Not during feed fetches where the feed already scoped media to per-entry content (trust the feed).
Cheap pre-checks (keep in code, don't spend AI tokens)
These checks are deterministic, free, and catch the overwhelming majority of obvious junk. Always run them before invoking this skill. If a pre-check drops an item, no AI call is needed.
- Tracking domains — URL host matches a known tracking/analytics domain (
px.ads.linkedin.com,t.co,www.facebook.com/tr,analytics.twitter.com,bat.bing.com). Drop with reasontracking domain: <host>. - Unsupported content-type — after HEAD/GET, content-type isn't in the uploadable set (
image/png|jpeg|gif|webp|svg+xml|avif,video/mp4|webm). Drop with reasonunsupported type. - Size bounds — body < 5 KB (tracking pixels, spacers) or > 10 MB (won't upload anyway). Drop.
- Streaming embeds — YouTube, Vimeo, Loom URLs are kept as
type: "video"references without downloading. Never route through R2 upload or this skill. - ETag / content hash seen before — if the R2 key derived from content hash already exists, reuse it and skip reclassification.
Everything else — the ambiguous middle where URL patterns overlap between chrome and content — goes through the skill.
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.
- 9d ago First seen · 91 lines · 57 tokens per session scan A 3e432218b34e
classify-media-relevance is a skill published in the GitHub repository buildinternet/releases (6 stars, last pushed 6d ago), licensed Apache-2.0. It adds 57 tokens to every session and 1,309 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 2026-08-31.
Other skills, from other repositories
seedance-pipeline
This skill should be used when the user asks about Seedance 2.0 workflow operations, API planning, BytePlus ModelArk, Dreamina/Jimeng surfaces, provider/router APIs, China-facing surfaces, ComfyUI, post-production, stitching, batch workflow, or integration planning.
presentation
Turn a tech-spec directory into an interactive, marketing-grade web presentation — built so engineers understand the design, the reader is convinced of the why, and the result is shareable in public. Use when someone wants a spec turned into a deck.
slide-deck-generator
Generate a real PowerPoint .pptx deck from a Markdown outline. Picks slide layouts (title / content / closing) based on the outline structure, applies a brand accent color, optionally adds speaker notes. Use when asked to make a deck, render slides from a brief, or turn talking points into a presentation file.
avatar-multi-scene
Use when someone wants the same person hosting several clips — multi-segment UGC, comparison reels, or mixed speaking and animated scenes with continuity.
p-image-ideogram
Use when photo generation needs more control — photoreal results, text in the image, or structured JSON with hex colors and bounding boxes. Simpler photo generation, edits, and video use other skills in the suite.
illustrated-story-reel
Use when someone wants a slideshow story with narration or music — picture-book illustrated frames with Ken Burns or gentle p-video motion.