Research crawl
1) crawl-map the target. 2) fetch-safe top URLs. 3) ask-json extract facts. 4) write a brief with citations. Stop when token-meter hits budget.
Catalog / workflows
Workflow playbooks
Repeatable agent loops. Run them in Cursor with credits from OneAIPass or API keys from cheaperllm.
Ship and review
Implement → diff-pack → prompt-lint on any new prompts → run tests → open PR via GitHub MCP. Strong model only on the review turn.
Cost-aware model routing
Classify intent with a cheap model; escalate planning and multi-file edits to a strong route (Claude via cheaperllm). Log every hop with token-meter.
Domain launch
Brainstorm names → domain-check → compare renewal on Agora / TLD Bazaar / BareDomains → register → drop static site + llms.txt.
MCP bootstrap
Install filesystem + fetch + memory. Attach anymodl or cheaperllm MCP. mcp-probe to verify. Commit mcp.json from the starter pack.
Continuous agent loop
Plan → act → observe → tape → decide. Full README in the workflow-agent-loop pack. Cap iterations and spend before starting.