Release v0.0.2

Fixed

• A fresh install writes the board's baseUrl (the hosted platform URL unless you override it), so the loop connects to the board without setting BASE_URL; the connect skill writes organizationId, the loop also reads orgId, and a loop missing credentials prints which fields are missing (#34)
• A slice that keeps failing is set to Blocked with a board comment giving the last failure after a configurable number of attempts (3 by default), a hung turn is stopped after a turn timeout (60 minutes by default), two loops on one board no longer build the same slice, and the loop sets each slice's final status itself (#36)
• The agent builds from the slice's full definition: before each build the loop fetches its elements, fields and specifications into slice.json, and each build turn names the slice to build (#37)

Removed

• The Ollama runner (ralph-ollama.js), which could not read the loop's tasks or build a planned slice; building requires Claude Code (#35)
• The undocumented shell loop (ralph.sh and lib/agent.sh), which also skipped every permission check

Security

• The loop no longer runs Claude with --dangerously-skip-permissions: file edits are accepted inside the project only, and the agent may run just the commands the kit needs (dotnet build/test/restore, branch-safe git, yarn install/build/lint/test, curl). Extend the list with allowedTools in the kit config or RALPH_ALLOWED_TOOLS; a refused command is reported as a blocked slice (#31)
• The board token is no longer written into the agent's prompt. It reaches the skills and the board MCP server through the EVENTMODELERS_TOKEN environment variable; rotate any token used with an earlier version (#32)
• The code export server listens on 127.0.0.1 only, answers browser requests from the board's origin only, and requires a per-install secret (codeExportSecret in .build-kit/.eventmodelers/config.json) as Authorization: Bearer <secret> or X-Export-Secret (#33)