mirror of
https://github.com/onestardao/WFGY.git
synced 2026-04-28 03:29:51 +00:00
7.3 KiB
7.3 KiB
Prompt Injection — Guardrails and Fix Patterns
🧭 Quick Return to Map
You are in a sub-page of Safety_PromptIntegrity.
To reorient, go back here:
- Safety_PromptIntegrity — prompt injection defense and integrity checks
- 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 desk within a ward.
If you need the full triage and all prescriptions, return to the Emergency Room lobby.
A focused guide to handle prompt injection attacks in RAG, agents, and orchestration.
Use this page when injected text hijacks your instructions, bypasses schema, or makes the model ignore contracts.
When to open this page
- Responses contain leaked system prompt or hidden instructions.
- Model obeys malicious user text like “ignore above and do X”.
- Citations vanish after injection payload.
- JSON / tool schema is broken by arbitrary free text.
- Memory or context keys rewritten by injected content.
Open these first
- Visual map and recovery: RAG Architecture & Recovery
- Retrieval traceability: retrieval-traceability.md
- Data schema contract: data-contracts.md
- Role boundary checks: role_confusion.md
- Memory fences: memory_fences_and_state_keys.md
Core acceptance
- ΔS(question, retrieved) ≤ 0.45 even with injection attempts.
- λ remains convergent across 3 paraphrases, does not flip under “ignore above” payloads.
- Schema lock: JSON/tool calls validate against fixed schema.
- Coverage ≥ 0.70 of target section even under noisy injection.
Fix in 60 seconds
-
Detect abnormal ΔS drift
- Compute ΔS(question, retrieved). If injected phrase raises ΔS ≥ 0.60, isolate payload.
-
Enforce contracts
- Wrap retriever and reasoner outputs in data-contracts.md.
- Reject free text outside schema.
-
Apply fences
- Lock system vs user roles (role_confusion.md).
- Use memory hash keys (memory_fences_and_state_keys.md).
-
Verify stability
- Re-run with paraphrase probes. Injection should not flip λ or erase citations.
Typical injection payloads → exact fix
| Payload type | Symptom | Fix |
|---|---|---|
| Ignore-all override | Model discards earlier rules | role_confusion.md + schema locks |
| Citation erasure | No references, only free text answer | retrieval-traceability.md, data-contracts.md |
| Tool hijack | JSON field replaced with instruction text | json_mode_and_tool_calls.md |
| Role swap | User prompt injected as “system” | role_confusion.md |
| Memory overwrite | Past state or keys corrupted | memory_fences_and_state_keys.md |
Copy-paste probe prompt
System: WFGY firewall active.
User input: {question}
Check:
1. Did retrieved snippet keep citations?
2. Did ΔS(question,retrieved) ≤ 0.45?
3. Did λ stay convergent under paraphrase?
4. Did JSON/tool call respect schema?
If any fail, return the failing layer + fix page.
🔗 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 + <your question>” |
| TXT OS (plain-text OS) | TXTOS.txt | 1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly |
Explore More
| Layer | Page | What it’s for |
|---|---|---|
| ⭐ Proof | WFGY Recognition Map | External citations, integrations, and ecosystem proof |
| ⚙️ Engine | WFGY 1.0 | Original PDF tension engine and early logic sketch (legacy reference) |
| ⚙️ Engine | WFGY 2.0 | Production tension kernel for RAG and agent systems |
| ⚙️ Engine | WFGY 3.0 | TXT based Singularity tension engine (131 S class set) |
| 🗺️ Map | Problem Map 1.0 | Flagship 16 problem RAG failure taxonomy and fix map |
| 🗺️ Map | Problem Map 2.0 | Global Debug Card for RAG and agent pipeline diagnosis |
| 🗺️ Map | Problem Map 3.0 | Global AI troubleshooting atlas and failure pattern map |
| 🧰 App | TXT OS | .txt semantic OS with fast bootstrap |
| 🧰 App | Blah Blah Blah | Abstract and paradox Q&A built on TXT OS |
| 🧰 App | Blur Blur Blur | Text to image generation with semantic control |
| 🏡 Onboarding | Starter Village | Guided entry point for new users |
If this repository helped, starring it improves discovery so more builders can find the docs and tools.