Skip to content
familiarfamiliarfamiliar

14. Known limits, stated plainly

  1. Rules are instructions, not enforcement. Every CORE-* rule depends on the model following it. mechanical-enforcement is available on Claude Code, Codex and Gemini CLI, and the guards are not part of the core download. The guards help enforce some rules, but they are not a guarantee. They match command text, so a short script run through an interpreter gets past them. The true claim is therefore “the guards catch the common cases where the agent supports hooks”, never “familiar prevents X”.
  2. Nothing verifies document-tools in advance. No capability has an install-time check, and the capability table is written by hand. The truth is whatever the live --health-check reports on the computer that runs the session. On an agent with no Python, deck and document builds do not work, and familiar says so.
  3. deck ships no template and no shared library. The only source is a template for this project that you place in 1-Inputs/Templates/. Without one, deck reports that it is unavailable. It does not invent a brand or a neutral master (§9.5).
  4. There is no calendar link. Date changes go to 4-Trackers/timeline.md and memory, as calendar-rules.md describes. familiar then tells you what to put on your own calendar.
  5. Concurrency is detection, not control. CORE-STALE-WRITE catches a second writer that follows the protocol. familiar supports one active writer at a time.
  6. user-invocation-control is an instruction only on Gemini CLI. Claude Code frontmatter and Codex settings enforce it. Gemini CLI relies on registry rule 3.
  7. The bootstrap has no room left. It is at its 500-word ceiling, and AGENTS.md holds a copy made by hand. Any addition needs a cut.
  8. Agent features change on their own schedule. A change in an agent’s file formats can require updates to the shipped subagent files and the documentation. Agents other than these three have no capability record (§10.3).