Add agent skill for driving the CLI from coding agents

Merged
KA
19h ago
+227-0

Why

Coding agents (Claude Code, Codex, Cursor, OpenCode) are now the main way people pick up dev tools. rae has a scripted JSON CLI but nothing that tells an agent how to use it.

What

  • skills/reverse-api-engineer/SKILL.md: install check, reuse of earlier runs (only when show reports a client, and a HAR before using engineer), --dry-run preflight including the chrome-mcp auto-connect check, background capture with --json-stream, verifying the client with run, iterating with engineer on HAR-backed runs, handing it over.
  • references/: setup and config keys, providers and logged-in sites, manual-mode handoff, troubleshooting by error_kind.
  • README install line (npx skills add kalil0321/reverse-api-engineer) and changelog entry.

Verification

  • npx skills add <repo> --list discovers the skill.
  • Fresh Claude Code sessions with only the skill installed:
    • rae 0.14.0 from PyPI, Node 24, YC company directory: preflight, background capture, generated client, live run with real filtered data.
    • #148 build, Node 26, Hacker News search with "don't use my Chrome" and "add a typed model by iterating on the same capture": the agent found an earlier HAR-backed run, verified its client, added the model with engineer ($0.95), and checked live pagination without touching the user's Chrome.

Out of scope

  • The chrome-mcp default, auto-connect preflight and engineer rejection of HAR-less runs ship in #148; the skill reads correctly on 0.14.0 too, but those parts only apply after #148 is released.
  • An MCP server so the host agent writes the client itself, and a setup command.