@AGENTS.md

Claude Code notes for WikiMem

Everything about what WikiMem does, its tools and what not to do is in AGENTS.md, imported above. This file keeps only what is specific to Claude Code.

The harness in this repository

• .claude/settings.json: two PreToolUse hooks. block-dangerous-commands.cjs stops destructive shell commands; protect-secrets.cjs stops reads and writes of key files.
• .claude/agents/: the sub-agents listed in AGENTS.md. They research and report; you make the change.
• .claude/skills/, .claude/rules/, .claude/commands/: general working methods inherited from the fleet's template. WikiMem's own skills are in skills/.
• .claude-plugin/: the Claude Code plugin (plugin.json, marketplace.json). It ships skills/, the two WikiMem sub-agents, the /wikimem-* commands in templates/claude-commands/, the vault hook in hooks/hooks.json and the MCP server in .mcp.json. Check it with claude plugin validate ..

MCP servers

• .mcp.json holds only WikiMem's own server, so it is safe for anyone who installs the plugin.
• .mcp.dev.json holds the servers used while developing (GitHub, Context7, a memory graph over memory/, an Obsidian server over brain/). Load them with claude --mcp-config .mcp.dev.json. The GitHub server reads GITHUB_PERSONAL_ACCESS_TOKEN from your environment.

Operating model

Act, then verify. Run the gate (pnpm build && pnpm test && pnpm eval) before you say a change is done, and test a UI change in the running app (wikimem serve, then node scripts/vibe-audit.mjs <tag>). When a correction teaches a rule, add it to memory/LEARNINGS.md.