chimeralang-mcp: Instructions file for Claude Code

ChimeraLang MCP — Project Memory

Token-saving policy (always active)

The canonical routing rules — which chimera_* tool to invoke for each prompt pattern, calling conventions, mode selection, worked examples, and skip conditions — live at:

• .claude/skills/chimera/SKILL.md — for Claude Code (auto-loaded as a skill)
• AGENTS.md — for Codex (same routing matrix at repo root)

Both files are kept in sync via tests/test_skill_integrity.py. Treat the SKILL.md routing matrix as binding for non-trivial tasks; skip chimera tools for prompts < 200 chars with no attachments.

Quick reference

Situation	Tool
Single large doc/code blob	chimera_optimize
Long conversation history	chimera_compress
Both docs + messages	chimera_fracture
Build/test/install logs	chimera_log_compress
Stable system blocks for an Anthropic SDK call	chimera_cache_mark
End-of-session totals	chimera_session_report

For everything else (reasoning lane, glyph lane, long tail, telemetry reactions), see .claude/skills/chimera/SKILL.md.

Hook coverage

The token tools fire automatically via .claude/settings.json:

• SessionStart — emits the policy as additionalContext.
• UserPromptSubmit — auto-compresses prompts > 800 chars.
• PreToolUse — warns on duplicate tool calls (dedup hit) and on Edit/Write/Bash payloads > 4000 chars.
• PostToolUse — records every non-chimera tool call into the dedup cache and nudges when responses > 2000 chars.
• Stop — emits a session totals summary (tokens saved, dedup hits) when the agent stops responding.

---

Karpathy Guidelines

Behavioral guidelines to reduce common LLM coding mistakes.

Tradeoff: These guidelines bias toward caution over speed. For trivial tasks, use judgment.

1. Think Before Coding

Don't assume. Don't hide confusion. Surface tradeoffs.

Before implementing:

• State your assumptions explicitly. If uncertain, ask.
• If multiple interpretations exist, present them - don't pick silently.
• If a simpler approach exists, say so. Push back when warranted.
• If something is unclear, stop. Name what's confusing. Ask.