Install https://trusty-kernel-dp3j.here.now/one-api-skill.zip and read https://trusty-kernel-dp3j.here.now/llms.txt. The gateway is https://one-api.161-35-251-195.sslip.io. Use the agent key provided privately in ONE_API_TOKEN; never put it into chat or source files. Discover live capabilities, choose the tools for my goal, work within the existing authorized trial, and save and verify the result. Do not assume inventory-only or missing connections are usable.
Give your agent
the tools to get it done.
Browsers, scraping, Treg research, enrolled Macs and saved progress. One private API and one portable skill, ready for the agents you trust.
Connect your agent
Give it the skill and a private, revocable agent key.
- Install the skillUnzip it into your agent's skills folder, or use the Python client directly.
- Provision its private keyThe owner creates a named key. Supply it through your agent's secret store or private connection file.
- Discover, then give it a goalThe live catalog reports what can execute, what needs a connection and what is only upstream inventory.
export ONE_API_URL=https://one-api.161-35-251-195.sslip.io # Set ONE_API_TOKEN privately through your runtime secret store. python3 client.py discover
MCP endpoint: https://one-api.161-35-251-195.sslip.io/mcp. Set the Authorization header from the agent's secret store. Download a connection template.
A useful tool for each step
Execution status always comes from authenticated discovery.
Treg research
Search the catalog, check prices and use four enabled web research and email enrichment routes. The OS includes four optional workflows.
treg.catalog.* · treg.web.* · treg.email.*Shared trial limits · UGC planning only, no generationYour native Macs
Reach the same four enrolled Mac agents through outbound HTTPS, preserving their native tools, sessions and job identity.
fleet.status · fleet.callSeparate native billing · private owner signature requiredCloud browsers
Render a public page in a fresh Browser Use Cloud browser. Collect text and links, then stop it and account for its cost.
browser.read · browser.resultSame-origin pages · 1-minute expiry · shared concurrencyYour own scrapers
Save and version extraction code. Test a sample, retrieve with free HTTP, Firecrawl or Apify, and inspect the records and logs.
scraper.save · test · run · resultCustom JavaScript · cloud retrieval · local extractionSearch & read
Search through the existing Scrapebox connection. Read public pages through BrowserBox's free fetch lane.
web.search · web.fetch · web.scrapeQuote first when needed · shared trial ledgerGoals & memory
Track acceptance criteria, tool receipts and versioned progress. Your calling agent plans the work and checks the result.
goal.* · memory.* · trial.statusSaved evidence · stable request IDs · no hidden workerGoal → tools → checked result
The productivity loop travels with the skill.
Discover
Inspect enabled tools and the authorized budget.
Act
Use the smallest useful tool with a stable request ID.
Verify
Check the actual records, source and completion criteria.
Save
Store evidence and progress; continue from saved state.
Unknown outcomes stay uncertain. A retry with the same request ID returns the existing job. Paid work reserves its maximum before dispatch. Another goal does not reset the allowance.
Everything for the handoff
These downloads contain no credentials or private workspace data.