npx skills add ...
npx skills add peter-duffy95/open-images-for-free-use-skill --skill open-images-for-free-use
Find images you can confidently use for free — for websites, documents, presentations, UIs, blog posts, marketing assets, or anything else — without ever worrying about licensing. This skill pulls from a curated repository of pre-vetted images, pinned to a specific human-reviewed revision, and confirms each image's license — read from the image's own embedded metadata — before handing it over, so the licensing question is already answered by the time you get the URL. Triggers on any request involving an image, photo, illustration, banner, hero, or visual asset, including phrases like 'find me an image of...', 'I need a photo of...', 'get a picture for...'. Also use this when building UIs, articles, or content where a stress-free, freely-licensed image is needed.
npx skills add peter-duffy95/open-images-for-free-use-skill --skill open-images-for-free-use
This is the one-stop way to source an image for any use case — UI hero, blog post, presentation, document, marketing asset — without ever wondering whether it's safe to use. It draws from a curated repository of pre-vetted images and checks each one's license against bundled, machine-readable license terms. By the time the skill hands back a URL, the licensing question is already answered.
There are two stages: pick an image and verify its license. Both are required — even though the source is curated, the license check is the contractual ground truth, and it's what makes this skill safe to trust.
Why this is safe to install (for reviewers):
- No third-party free text reaches the agent at runtime. The catalog (
./index.json) is shipped inside the skill and reviewed like any other bundled file. The only thing fetched at runtime is the selected image's bytes, and the only value extracted from them is a license identifier that is immediately allow-listed against a fixed set of eight known IDs — it cannot carry arbitrary text into context, and it can never name a path. Raw image bytes /stringsoutput are never printed.- Fetches are pinned to an immutable commit SHA (
e40900d), not a moving branch. The bytes at a given SHA cannot change, so what was reviewed is what ships — a movingmainis exactly the "external content changes under you" risk that pinning removes.- The license verdict is grounded, not assumed. It comes from the image's own embedded metadata (the source's TOS makes that the authoritative declaration), resolved to a license file shipped inside this skill. The images at the pinned revision were individually license-reviewed by the author.
- Release-time invariant: when you bump the pinned SHA to add images, re-ship the matching
./index.jsonfrom that SHA and re-review both the new licenses and the new catalog entries — the "pre-vetted" guarantee is only honest for a SHA whose contents you have actually checked.
The catalog is shipped with this skill as ./index.json — it is not fetched at runtime. It is a JSON array; each entry has a name (the image filename) and a keywords array describing the image. Because it travels inside the skill, it is reviewed content, just like the license files — no outsider-authored free text enters the agent's context at runtime.
The catalog is treated as a suggestion only: it tells the skill which image to fetch, but the license verdict comes from the image file itself, and that verdict is allow-listed (see below) before it can affect anything. So no value the catalog supplies is ever used as a path or trusted blindly.
Read the local catalog directly — do not pre-filter with a keyword query. Look at the full list, then pick the entry whose name and keywords most closely match the user's intent (consider what the image will be used for — hero banner, blog header, illustration, portrait — not just the literal subject). Use judgment, not substring matching.
If nothing in the catalog feels like a good fit, pick the closest available option and flag to the user that the match is approximate.
Once an entry is chosen, set image_name to its name value and run the verification block. It fetches the image from the pinned SHA, reads the license identifier embedded in the image's metadata, checks that identifier against a fixed allow-list of the eight known license IDs, and only then prints the matching license file shipped inside this skill. Run it from the skill's own directory so the relative ./licenses/ path resolves regardless of where the skill was installed:
The license file is the ground truth — never assume an image is free to use just because it appears in the catalog.
The bundled licenses are:
CC0-1 — Creative Commons Zero (public domain dedication)CC-BY-4 — Attribution onlyCC-BY-SA-4 — Attribution + ShareAlikeCC-BY-NC-4 — Attribution + NonCommercialCC-BY-NC-SA-4 — Attribution + NonCommercial + ShareAlikeCC-BY-NC-ND-4 — Attribution + NonCommercial + NoDerivativesCC-BY-ND-4 — Attribution + NoDerivativesGFDL-1 — GNU Free Documentation LicenseRead the resolved license file and check whether its terms are compatible with how the image will be used. Key considerations:
If no matching license file is found, or the license terms are incompatible with the intended use, go back to Stage 1 and pick a different image.
Return the verified image URL. If the license has attribution requirements, communicate them to the user so they can comply.