davidtoby Automation, Browser, and API Workflows
Use this umbrella when the task is to operate an external interface repeatedly or replace brittle UI actions with a more reliable API/tool path.
Core workflow
- Identify the interface: browser page, native macOS UI, interactive terminal app, MCP server, RTSP/ONVIF camera, or HTTP API.
- Prefer structured APIs/CLIs over visual clicking when available.
- If starting from browser automation, capture state, network traffic, and request/response shapes so the workflow can be converted to direct calls.
- Keep credentials out of logs and references; store only redacted examples.
- Build deterministic probes/scripts for repeated actions and verify outputs after each side effect.
Labeled playbooks
Browser automation
Use snapshots/selectors/click/type flows for one-off navigation. For repeated work, graduate to direct HTTP or a specialized CLI.
API learner
Record traffic, identify auth/session headers, replay with curl, minimize payloads, and document stable endpoints plus failure codes.
macOS UI automation
Use screenshot/element capture tools for apps that lack APIs. Verify focus and target window before sending keystrokes.
tmux and interactive CLI control
For long-lived or REPL-like tools, send keystrokes and scrape panes instead of running non-PTY commands that will hang.
MCP tool calling
List servers/tools, inspect schemas, then call with JSON arguments. Generate typed wrappers only after schemas are stable.
Camera capture
Probe stream URLs and capture short frames/clips with ffmpeg-compatible tools; avoid assuming camera availability.
Reference files
Exact CLI commands and session-specific endpoint notes are stored in references/from-*.md.