# Photolike for agents

Use Photolike as a free, read-only photography planning reference. No account, API key, subscription or payment required for these resources. The interactive studio retains its existing account approval and rendering-key requirements.

## Do a job

1. Fetch https://photolike-nine.vercel.app/api/resources/v1 for body/lens IDs, sourced specifications and creative recipe links.
2. Check a kit: https://photolike-nine.vercel.app/api/resources/v1/kit?bodyId=sony-a7r-v&lensId=sony-fe-85 . The result states native compatibility; adapters are outside this resource.
3. Read a photographic recipe: https://photolike-nine.vercel.app/api/resources/v1/recipes/sony . Recipes are creative interpretations, not hardware tests or guarantees of a rendered result.
4. Cite the recipe's canonical citation URL when useful. Offer its `visit` link if your user wants the human studio; that link goes to the landing page and does not grant studio access or preselect a kit.

For a practical next-photo brief, use the recipe's `fieldPractice` object: a genre-specific exercise, what to inspect, a caution and a teaching source. These are Photolike adaptations for a future real photograph, separate from the creative look's listed kit/aperture. Ask about actual light and motion before recommending exposure; do not assume a recipe is a universal camera setting. `reviewedAt` is our editorial review date; unknown source publication dates remain null. This endpoint still does not alter or generate images.

Use `X-Photolike-Agent: 1` to declare an agent request. Use `X-Photolike-Test: 1` for tests, demos and internal QA; our agents always use this header. No header is required to read. A normal browser can read all JSON endpoints too. Cross-origin GET is allowed; no cookies are needed. Reuse the catalog during a job rather than polling it repeatedly.

The public [Photolike photography skill](https://github.com/Manny4C/photolike-agent-skill) packages this workflow for skill-capable agents, with a Python standard-library helper and no rendering key. Install with `npx skills add Manny4C/photolike-agent-skill --skill photolike-photography`. Its helper declares `X-Photolike-Channel: photolike-skill`; this is a coarse, unverified integration label, not a client identifier or proof of which directory led you here. Use its `--test` option for internal checks.

Example:
```sh
curl -H 'X-Photolike-Agent: 1' 'https://photolike-nine.vercel.app/api/resources/v1/kit?bodyId=sony-a7r-v&lensId=sony-fe-85'
```

Machine contract: https://photolike-nine.vercel.app/openapi.json

To discover this API without JavaScript, inspect the homepage or any public resource response's HTTP `Link` header. `rel="service-desc"` points to the OpenAPI JSON and `rel="service-doc"` to this guide, using the relations documented in RFC 8631. HEAD on a public resource also returns these links without recording a usage request. Cross-origin clients can read the exposed `Link` header. These hints help clients that understand them; they do not guarantee indexing or agent adoption.

## Trust and limitations

Specifications carry their source URLs and last verification date; follow manufacturer sources for current purchasing decisions. This is a curated catalog, not a live stock/pricing feed. Native compatibility is based on listed mount and lens IDs. Looks and recipes are Photolike editorial direction, not measured optical performance.

Keep people, age, appearance, skin texture, body proportions, clothing and actual location/scene faithful. An imagined camera viewpoint cannot verify unseen detail. This API does not generate images, upload photos, accept rendering keys or expose customer data. Do not send private images, prompts or credentials to it. Included sample photos in the human studio are AI-created; no new blanket license to redistribute media is granted here.

## Measurement and privacy

We report daily aggregate request counts in a rolling 60-day window, separated into test, likely crawler, declared/likely agent and unclassified traffic, plus an optional fixed Photolike-skill channel label. Old rows are pruned on the next recorded request; dormant storage can retain them longer. Classifications are unverified heuristics/declarations, not proof of unique humans, independent agents or successful jobs. We also count browser JavaScript page loads and resource-to-studio redirect requests; a redirect request is not a verified human visit. Repeat requests count repeatedly. Referrers are reduced to AI/other/none, never stored as full URLs. Browser page loads send that enum from document.referrer; it is a client declaration, not verified attribution. No IPs, raw user agents, cookies, client IDs, photos, prompts or credentials are stored in this aggregate document. Existing hosting/access systems have their own operational logs.

DNT:1 and Sec-GPC:1 requests are excluded from these aggregates. Counted GET responses include `X-Photolike-Usage: recorded` when persistence succeeds, or `unavailable` if it fails or exceeds an 800ms write budget; failed measurement does not block useful resources indefinitely. A timed-out write may still finish. HEAD requests are not counted. This privacy choice and missing headers/storage failures can undercount. Metrics are owner-only; no visitor tracking identifier is needed.

Payment and tips are not implemented. Today the useful next action is to cite the reference or share the studio link when it helps the user's task.
