vector-memory-mcp: Command for Claude Code

Iron Rules

No-hallucination (CRITICAL)

NEVER output results without ACTUALLY calling tools. You CANNOT know task status or content without REAL tool calls. Fake results = CRITICAL VIOLATION.

No-verbose (CRITICAL)

FORBIDDEN: Wrapping actions in verbose commentary blocks (meta-analysis, synthesis, planning, reflection) before executing. Act FIRST, explain AFTER.

Max-interactivity (CRITICAL)

When $HAS_AUTO_APPROVE = false: MUST engage user with clarifying questions via AskUserQuestion tool. NEVER assume scope, depth, audience, or structure. Documentation is a COLLABORATIVE process — user defines WHAT, agent researches and writes HOW. When $HAS_AUTO_APPROVE = true: infer scope/depth/audience from $CLEAN_ARGS context. Skip clarifying questions. Proceed autonomously through all phases.

• why: Wrong assumptions about documentation scope = useless output + full rework. Interactive alignment is cheaper than rewrites. But -y flag means user trusts agent to make reasonable decisions autonomously.
• on_violation: If interactive: STOP and ask clarifying question. If auto-approve: infer from input and proceed.

Discovery-before-creation (CRITICAL)

ALWAYS search existing docs via brain docs CLI BEFORE creating new files. Flow: brain docs "{keywords}" → found? → READ existing → UPDATE. Not found? → apply aggressive-docs-search (3+ keyword variations). Still not found → CREATE new. NEVER create duplicate documentation for same topic.

• why: Duplicate docs diverge over time. One source of truth per topic. Updating existing is faster and preserves history.
• on_violation: Run brain docs first. Found → update. Not found after 3+ searches → create new.