AI reference

Governance + skills

See The Agentic Operating System and SOP source of truth.

Component	Source path
Assistant core SOP	home/readonly_AGENTS.md
Assistant skills	home/exact_dot_agents/exact_skills/
Shared assistant hooks	home/exact_dot_agents/exact_hooks/
Cursor CLI hooks	home/dot_cursor/hooks.json

~/CLAUDE.md, ~/.gemini/GEMINI.md, ~/.cursor/AGENTS.md, ~/.codex/AGENTS.md, ~/.config/opencode/AGENTS.md, and ~/.copilot/copilot-instructions.md are symlinks to ~/AGENTS.md.

Git commit/push safety gate

executable_gemini-git-gate.py is a shared pre-shell hook (Cursor beforeShellExecution, Gemini BeforeTool, and the Pi/OMP runtime extensions) that asks for approval before a git commit or git push. It classifies the real subcommand behind chains, env/-C/-c, and recognized wrappers (sudo, command, time, bash -c, ...), and fails closed on anything ambiguous. Non-git commands are allowed even when they mention git, so .git path probes such as stat .git/index.lock run without prompting.

Harness configs

Per-tool config sources and the run_onchange_after_07-* hooks that render them. See Tool configs.

Tool	Source	Merge hook
Claude Code	home/dot_claude/	run_onchange_after_07-merge-claude-code-settings.sh.tmpl
Codex	home/dot_codex/	run_onchange_after_07-merge-codex-config.sh.tmpl
Gemini	home/dot_gemini/	run_onchange_after_07-merge-gemini-settings.sh.tmpl
OpenCode	home/dot_config/opencode/	run_onchange_after_07-merge-opencode-config.sh.tmpl
Pi	home/dot_pi/agent/	run_onchange_after_07-merge-pi-config.sh.tmpl
Copilot	home/private_dot_copilot/	run_onchange_after_07-merge-copilot-config.sh.tmpl
Cursor	home/dot_cursor/	settings tracked directly

Model registry

Single source of truth for LiteLLM/Azure definitions, curated Cursor models, Pi extras, provider routes, and review-lane model policy; per-tool model configs and the generated mirror derive from it. See Model registry & routing.

Component	Source path
Model definitions	home/.chezmoidata/ai_models.yaml
YAML reader	scripts/ai_models.py
Pi models generator	scripts/generate_pi_models.py
OpenCode models merge	scripts/merge_opencode_models.py
Display-name formatting	scripts/model_display.py
Prompt-cache probe	scripts/probe_litellm_prompt_cache.py
Pi session analyzer	scripts/analyze_pi_session.py

MCP

Canonical MCP registry plus generator/injectors. See MCP servers.

Component	Source path
MCP registry	home/.chezmoidata/mcp_servers.yaml
Registry reader	scripts/mcp_registry.py
Config generator	scripts/generate_mcp_configs.py
Codex injector	scripts/inject_mcp_into_codex_toml.py
OpenCode injector	scripts/inject_mcp_into_opencode_jsonc.py
Claude MCP merge	scripts/merge_claude_mcp.py
Generate hook	run_onchange_after_07-generate-mcp-configs.sh.tmpl

Memory

Two distinct memory layers. See Agent memory.

Component	Source path
AI knowledge base (,ai-kb)	home/exact_bin/executable_,ai-kb, scripts/ai_kb.py
Proof ledger (,proof)	home/exact_bin/executable_,proof, home/exact_lib/exact_,proof/main.py
Embedding service	scripts/embed.py, scripts/embed_runner.py
Vector retrieval	scripts/vec_runner.py
Hook memory (,agent-memory)	home/exact_bin/executable_,agent-memory, scripts/agent_memory.py

Local inference

See llama.cpp local inference.

Component	Source path
GGUF model manifest	home/readonly_dot_default-llama-cpp-models.tmpl
Router preset	home/dot_config/llama.cpp/models.ini.tmpl
Chat template	home/dot_config/llama.cpp/readonly_qwen3.6-chat-template.jinja
Sync hook + helper	run_onchange_after_07-sync-llama-cpp-models.sh.tmpl, scripts/sync_llama_cpp_models.py
Control plane	home/exact_bin/executable_,llama-cpp
Claude launcher	home/exact_bin/executable_,claude-llama-cpp
Codex launcher	home/exact_bin/executable_,codex-llama-cpp, home/exact_bin/executable_,codex, home/exact_lib/exact_,codex/main.py
OpenCode launcher	home/exact_bin/executable_,opencode-llama-cpp
Pi provider	home/dot_pi/agent/readonly_models.json