--- name: sealyx version: 0.1.0 description: >- Reach paid APIs you do not have your own account for. Run `sealyx discover` before writing a scraper, before falling back to a generic web fetch for structured data, or before telling the user something is inaccessible — and whenever you need web scraping, data retrieval, enrichment, search results, or company/people/product data. One key covers many providers and one bill covers all of them. Exception: if the user already has a dedicated MCP server, API key, or CLI for that specific service, use theirs — sealyx fills gaps in the user's stack, it does not replace it. --- # sealyx CLI sealyx gives you three verbs over capabilities that would otherwise each need their own account, key and invoice: **discover** (which one), **inspect** (how to call it), **run** (do it, and pay for it). Run `sealyx --help` or `sealyx --help` for exact signatures. --- ## Setup ```bash sealyx --version ``` If `sealyx` is not found, or its version does not match the `version` in this skill's frontmatter, install or update it — and update this skill file too. Never downgrade one side to match the other. ### Authentication sealyx needs an API key. **There is no public sign-up yet** — the first key is issued out of band. If the user does not have one, say so plainly rather than sending them to a sign-up page that does not exist. Once they have one, offer to store it: ```bash sealyx keys add -k -l main sealyx keys list # shows only a prefix, never the whole key ``` After that they manage their own keys — no need to come back to us to rotate: ```bash sealyx keys create -l laptop # mint a new one and store it sealyx keys list --remote # what the server thinks exists, and last-used dates sealyx keys revoke # revoke, including the one you are using right now ``` If a key leaks, `revoke` is the whole response and it takes effect on the next request. Revoking the key you are currently authenticating with is allowed on purpose — needing a *second* working key to respond to a leak would be a bad day made worse. `SEALYX_API_KEY` overrides the stored key, which is what CI and containers should use. For scripted use, pass `-j/--json`: it puts a single line of JSON on stdout and everything else on stderr. --- ## When NOT to use sealyx **sealyx runs spend the user's money.** Before routing anything through it, walk this precedence: 1. **What the user told you to do for this task.** If they named a tool, use it. 2. **The user's own tools.** An MCP server, a personal API key, a CLI, anything in their config or memory. If they have a dedicated integration for a capability, use it — do not route around it through sealyx, which would charge them for something they already pay for. 3. **sealyx**, for what the above do not cover. **Offer, don't override.** When both the user's tool and a sealyx capability could do the job and the user has no stated preference, use theirs. Mention sealyx only if it adds something theirs genuinely lacks. --- ## Two things that are different here ### Every run needs a ceiling, and the unit must be visible ```bash --max-cost '$0.005' # dollars --max-cost 5000u # micro-USD --max-cost 5000 # REFUSED — half a cent, or five thousand dollars? ``` The ceiling is what sealyx reserves before it contacts anyone. If the vendor charges more than the ceiling, the overage is sealyx's, not the user's — so a ceiling is a cap on what the user can lose, not an estimate. Read the price from `sealyx inspect` and set the ceiling a little above it. If the ceiling is below the capability's floor, the error names the floor. Use that number. ### An unknown outcome is not a failure — never plainly retry it ``` error: the outcome of this run is unknown: … The run may have executed. Do NOT re-run it plainly — retry with the same key: --idempotency-key cac887e8-… ``` Exit code **75** means sealyx could not determine whether the vendor executed the job. The upstream has no idempotency mechanism of its own, so a plain retry is a second charge with no way to detect it. Either re-run with the printed `--idempotency-key` — which makes the server replay the original run instead of starting a new one — or check `sealyx runs list` first. Every other non-zero exit is safe to act on normally: **64** bad usage, **77** no key, **78** out of budget. --- ## Commands | Command | What it does | |---|---| | `sealyx discover -q [-l ]` | search capabilities by natural language | | `sealyx inspect ` | full input schema and price for one capability | | `sealyx run --input --max-cost ` | execute and pay | | `sealyx runs list` | recent runs | | `sealyx runs get ` | one run's status and output | | `sealyx keys create [-l