Documentation
¶
Overview ¶
Package pipeline provides the deterministic observation extraction pipeline. Level 0 processing: no LLM required, rule-based classification and extraction.
Index ¶
- func ClassifyEvent(toolName, toolInput, toolResult string) models.ObservationType
- func ExtractConcepts(toolName, toolInput, toolResult string) []string
- func ExtractFacts(toolName, toolInput, toolResult string) []string
- func ExtractFilePaths(toolInput, toolResult string) []string
- func FormatEmbeddingText(obs *models.Observation) string
- func GenerateTitle(toolName, toolInput string) string
- func ShouldSkipTool(toolName string) bool
- func ShouldSkipTrivial(toolName, toolInput, toolResult string) bool
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func ClassifyEvent ¶
func ClassifyEvent(toolName, toolInput, toolResult string) models.ObservationType
ClassifyEvent determines the observation type from a raw tool event. Rule-based classification achieving ~80% accuracy vs LLM.
func ExtractConcepts ¶
ExtractConcepts extracts keyword-based concepts from the event. Returns deduplicated concept tags.
func ExtractFacts ¶
ExtractFacts extracts diff-based facts from tool events.
func ExtractFilePaths ¶
ExtractFilePaths extracts file paths from tool input and output.
func FormatEmbeddingText ¶
func FormatEmbeddingText(obs *models.Observation) string
FormatEmbeddingText creates compact text for embedding from an observation.
func GenerateTitle ¶
GenerateTitle creates a template-based title for the observation.
func ShouldSkipTool ¶
ShouldSkipTool checks if a tool should be skipped for observation extraction. Exported version of the filter from sdk/processor.go.
func ShouldSkipTrivial ¶
ShouldSkipTrivial checks if the operation is too trivial to process. Exported version of the filter from sdk/processor.go.
Types ¶
This section is empty.
Source Files