Unverified50% confidenceEventExact time
At hour 27 the author discovered the system prompt never entered the hash, because the system message was injected by a thin client wrapper reading an environment variable, causing two different prompts to produce the same cache key.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe fix for the system prompt not entering the hash: explicitly assert that the first item in the message list must be the system role, and write the byte hash of the prompt file into the cache key.70% similarUnverifiedWhether prompt caching hits depends on byte-for-byte consistency of the request prefix; if even one character at the beginning differs, the entire cache is invalidated.64% similarVerifiedThe root cause of Prompt Injection is that large models cannot structurally distinguish between "system instructions" and "user data," as both are presented as natural language text at input time.62% similarVerifiedThe root cause of prompt injection lies in the fact that LLM architecture encodes system prompts, user inputs, and externally retrieved content uniformly into the same continuous token sequence, making it impossible for the model to distinguish trusted instructions from untrusted data at the semantic level.61% similarUnverifiedWhen the request body comes from a file stream or generator, the stream pointer moves to the end after the first read, making it impossible to read the same content again during a retry.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1066603API
curl https://kongchang.com/api/v1/knowledge/claims/1066603MCP
get_claim(id=1066603)