| 1 | package anthropic |
| 2 | |
| 3 | import ( |
| 4 | "slices" |
| 5 | |
| 6 | "reasonix/internal/sessioncontext" |
| 7 | ) |
| 8 | |
| 9 | func sessionContextTextBlocks(content string) []contentBlock { |
| 10 | parts := sessioncontext.SplitBlocks(content) |
| 11 | blocks := make([]contentBlock, 0, len(parts)) |
| 12 | for _, part := range parts { |
| 13 | blocks = append(blocks, contentBlock{Type: "text", Text: part.Text}) |
| 14 | } |
| 15 | return blocks |
| 16 | } |
| 17 | |
| 18 | // markPromptCacheBreakpoints marks stable system/tools, the latest valid |
| 19 | // session context, and the request tail. Native Anthropic permits four cache |
| 20 | // breakpoints; Reasonix uses at most three and keeps the default five-minute TTL. |
| 21 | func markPromptCacheBreakpoints(system []textBlock, tools []anthTool, messages []anthMessage) { |
| 22 | if n := len(system); n > 0 { |
| 23 | system[n-1].CacheControl = ephemeral() |
| 24 | } else if n := len(tools); n > 0 { |
| 25 | tools[n-1].CacheControl = ephemeral() |
| 26 | } |
| 27 | |
| 28 | contextMessage, contextBlock := -1, -1 |
| 29 | for i := len(messages) - 1; i >= 0 && contextMessage < 0; i-- { |
| 30 | for j := range slices.Backward(messages[i].Content) { |
| 31 | block := messages[i].Content[j] |
| 32 | if block.Type == "text" && sessioncontext.IsContent(block.Text) { |
| 33 | contextMessage, contextBlock = i, j |
| 34 | break |
| 35 | } |
| 36 | } |
| 37 | } |
| 38 | if contextMessage >= 0 { |
| 39 | messages[contextMessage].Content[contextBlock].CacheControl = ephemeral() |
| 40 | } |
| 41 | if n := len(messages); n > 0 { |
| 42 | if k := len(messages[n-1].Content); k > 0 { |
| 43 | messages[n-1].Content[k-1].CacheControl = ephemeral() |
| 44 | } |
| 45 | } |
| 46 | } |
| 47 |