walki 0.2.0

Local coordination protocol for AI agents. Let agents debate, reach agreements, and produce reviewable decisions through Markdown files in your repo.

0.2.0

• MCP server: Walki is now usable as an MCP tool by any MCP-compatible agent (opencode, Claude Desktop, etc.)
  • walki_open_channel - Create a new debate channel
  • walki_read_channel - Read channel messages
  • walki_post_message - Append a message to a channel
  • walki_propose_decision - Propose a decision in a channel
  • walki_get_status - Get workspace or channel status
  • walki_close_channel - Close a debate
  • walki_promote_to_sdd - Promote a decision to sdd-ai
• Permission enforcement via MCP: validates agent roles, closed channels, and turn limits
• STDIO and HTTP transport support for MCP server
• New binary: walki-mcp for running as an MCP server

0.1.1

• Fix ChannelStatus.toYamlValue() to properly convert camelCase to kebab-case (needsHuman → needs-human)
• Fix PermissionEngine.validateMessage to use agent ID instead of role for message counting
• Add comprehensive test suite: 131 tests covering all core modules
• Update README with agent roles, full CLI reference, and usage examples

0.1.0

Initial release of Walki - local coordination protocol for AI agents.

Features

• CLI: 12 commands for managing agent debates

  • walki init - Initialize .walki/ workspace
  • walki agent add/list - Manage agent identities and roles
  • walki debate - Create debate channels in Markdown
  • walki say - Append messages to channels
  • walki read - Read channel contents
  • walki status - Show workspace or channel status
  • walki summarize - Generate structured summary
  • walki close - Close a debate with a status
  • walki promote - Promote decisions to sdd-ai
  • walki doctor - Validate workspace integrity
  • walki rules add/list - Manage project rules
  • walki export - Export debates as Markdown or JSON

• Markdown-first channels: Debates are versionable, diffable Markdown files

• Protocol permissions: Role-based validation for agents (implementer, reviewer, owner)

• Hierarchical instructions: Load rules from global, project, domain, and channel levels

• Channel parser/formatter: Full round-trip Markdown parsing and generation

• sdd-ai integration: Detect sdd-ai directories and promote decisions

• Agent prompts: Auto-generated prompts per agent role for Codex, Claude, etc.