WFGY/ProblemMap/GlobalFixMap/Safety_PromptIntegrity/README.md

Safety & Prompt Integrity — Global Fix Map

🏥 Quick Return to Emergency Room

You are in a specialist desk.
For full triage and doctors on duty, return here:

• WFGY Global Fix Map — main Emergency Room, 300+ structured fixes
• WFGY Problem Map 1.0 — 16 reproducible failure modes

Think of this page as a sub-room.
If you want full consultation and prescriptions, go back to the Emergency Room lobby.

A hub to stabilize prompt-level safety and schema integrity across providers, agents, and eval flows.
Use this folder when failures look like jailbreaks, role confusion, or malformed tool calls.
Each page maps symptoms → root cause → structural fixes with measurable acceptance targets.

---

What this page is

• A practical checklist for anyone shipping LLM apps with tools, roles, or multi-agent setups.
• Each failure pattern links to its own guide with copy-paste guardrails.
• Works without infra changes — schema and prompt fixes only.
• Acceptance targets (ΔS, λ, coverage) are reproducible.

---

When to use

• Jailbreak attempts slip past normal filters.
• Prompts collapse schema or inject rogue tools.
• Tool calls drift into free text or JSON breaks.
• Role instructions misalign (system vs user vs assistant).
• Citations disappear or retrieval bypasses snippet contracts.
• Eval pipelines show high ΔS drift even when retrieval is correct.

---

Common failure patterns

Failure mode	What happens	Open this
Prompt Injection	Hidden instructions override your system prompt	prompt_injection.md
Jailbreaks / Overrides	User tricks model into ignoring rules	jailbreaks_and_overrides.md
Role Confusion	System / user / assistant boundaries collapse	role_confusion.md
Memory Fence Missing	State leaks across runs, no stable key	memory_fences_and_state_keys.md
JSON Drift	Tool calls malformed, fields missing	json_mode_and_tool_calls.md
Citation Lost	Answers skip snippet or no “cite-then-explain”	citation_first.md
Injection Defense Recipes	Ready-to-paste guardrails against common exploits	anti_prompt_injection_recipes.md
Tool Timeouts	Tool calls hang or return late	tool_selection_and_timeouts.md
Role Ordering	Wrong order breaks downstream eval	system_user_role_order.md
Template Gaps	Prompts inconsistent across agents	template_library_min.md
Eval Drift	No stable way to test safety fixes	eval_prompts_and_checks.md

---

Acceptance targets

• ΔS(question, retrieved) ≤ 0.45
• Coverage of cited section ≥ 0.70
• λ convergent across three paraphrases and two seeds
• No uncontrolled free-text execution in JSON or tool modes
• Citation-first enforced in ≥ 95% of eval runs

---

60-second fix checklist

1. Lock system / user / assistant role order.
2. Enforce citation-first and snippet schema.
3. Apply JSON fences + argument validation.
4. Add memory fences keyed by mem_rev and state_key.
5. Run eval prompts + probes before shipping.

---

🔗 Quick-Start Downloads (60 sec)

Tool	Link	3-Step Setup
WFGY 1.0 PDF	Engine Paper	1️⃣ Download · 2️⃣ Upload to your LLM · 3️⃣ Ask “Answer using WFGY + ”
TXT OS	TXTOS.txt	1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly

---

🧭 Explore More

Module	Description	Link
WFGY Core	WFGY 2.0 engine: full symbolic reasoning & math stack	View →
Problem Map 1.0	Initial 16-mode diagnostic framework	View →
Problem Map 2.0	RAG failure tree and modular fixes	View →
Semantic Clinic	Expanded catalog: injection, memory bugs, logic drift	View →
Semantic Blueprint	Layer-based symbolic reasoning & semantic mods	View →
Benchmark vs GPT-5	Stress test GPT-5 with WFGY reasoning suite	View →
🧙‍♂️ Starter Village 🏡	New here? Lost in symbols? Wizard will guide you	Start →

---

👑 Early Stargazers: See the Hall of Fame
⭐ WFGY Engine 2.0 is already unlocked. Star the repo to unlock more.