[MODEL] Claude doesn't know its own CLI flags — users shouldn't have to teach it --resume

The problem

Claude doesn't know basic facts about its own CLI. In a real session today, it took ~30 user turns to get the correct command to resume a session:

claude -r <full-uuid>

Two facts were wrong:

1. Flag is -r, not -c. Claude recommended claude -c <session-id>. Wrong — -c continues the most recent session and takes no ID argument.
2. Full UUID required. Claude recommended a short prefix (0aad55d4). Wrong — short prefixes fail with "not a UUID and does not match any session title."

Claude confidently gave the wrong command, then had to be asked repeatedly to verify it before finally getting it right.

The contrast

Claude has built-in skills for advanced workflows — cleaning up permissions, code review, deep research, frontend design, security audits. It can orchestrate multi-agent workflows and manage MCP server connections. But it doesn't know that --resume takes a full UUID, or that -c and -r are different flags. The advanced stuff works; the basics are missing.

Suggestion

Bake Claude Code CLI usage (flags, slash commands, exact syntax) into training data or the system prompt. The user should not need to:

• Manually save a memory file teaching Claude its own flags
• Spawn a research agent to look up --resume syntax
• Burn 30 turns correcting wrong answers about a 2-flag distinction

This is Claude's own product. It should know how it works.

Related

• #52556 — no obvious recovery path after closing terminal
• #38503 — easy resume after accidental close (VS Code)
• #26137 — --resume can't find sessions after unexpected shutdown