Documentation
¶
Overview ¶
Package parser extracts structured content from LLM responses.
Core types:
- Response: Contains structured data extracted from an LLM response
- CodeBlock: A fenced code block with language and content
- Parser: Extracts code blocks, JSON, YAML, sections, and lists
Example usage:
p := parser.NewParser()
resp := p.Parse(llmOutput)
// Access extracted code blocks
for _, block := range resp.CodeBlocks {
fmt.Printf("Language: %s\nCode:\n%s\n", block.Language, block.Content)
}
// Access parsed JSON
for _, data := range resp.JSONBlocks {
fmt.Printf("JSON: %v\n", data)
}
Convenience functions:
json := parser.ExtractJSON(response) code := parser.ExtractCode(response, "go") parsed := parser.Parse(response)
Index ¶
- Variables
- func ExtractCode(response, language string) string
- func ExtractJSON(response string) map[string]any
- func GetBlockedReason(content string) string
- func IsPhaseBlocked(content string) bool
- func IsPhaseComplete(content string) bool
- type CodeBlock
- type Marker
- type MarkerMatcher
- func (m *MarkerMatcher) AddTag(tag string)
- func (m *MarkerMatcher) Contains(content, tag string) bool
- func (m *MarkerMatcher) ContainsAny(content string) bool
- func (m *MarkerMatcher) ContainsValue(content, tag, value string) bool
- func (m *MarkerMatcher) FindAll(content string) []Marker
- func (m *MarkerMatcher) FindAllForTag(content, tag string) []Marker
- func (m *MarkerMatcher) FindFirst(content, tag string) (Marker, bool)
- func (m *MarkerMatcher) GetValue(content, tag string) string
- func (m *MarkerMatcher) Tags() []string
- type Parser
- func (p *Parser) ExtractAllCode(response string) []CodeBlock
- func (p *Parser) ExtractCode(response, language string) string
- func (p *Parser) ExtractJSON(response string) map[string]any
- func (p *Parser) ExtractJSONArray(response string) []map[string]any
- func (p *Parser) ExtractList(response string) []string
- func (p *Parser) ExtractNumberedList(response string) []string
- func (p *Parser) ExtractSection(response, title string) string
- func (p *Parser) ExtractYAML(response string) []map[string]any
- func (p *Parser) HasCodeBlock(response string) bool
- func (p *Parser) HasJSON(response string) bool
- func (p *Parser) Parse(response string) *Response
- type Response
Constants ¶
This section is empty.
Variables ¶
var ( // PhaseMarkers matches common phase completion markers. PhaseMarkers = NewMarkerMatcher( "phase_complete", "phase_blocked", ) // TaskMarkers matches task-level markers. TaskMarkers = NewMarkerMatcher( "task_complete", "task_blocked", "task_failed", ) )
Common marker matchers for orchestration systems.
Functions ¶
func ExtractCode ¶
ExtractCode is a convenience function for code extraction.
func ExtractJSON ¶
ExtractJSON is a convenience function for JSON extraction.
func GetBlockedReason ¶ added in v1.1.0
GetBlockedReason extracts the reason from a phase_blocked marker. Returns empty string if no blocked marker is found.
func IsPhaseBlocked ¶ added in v1.1.0
IsPhaseBlocked checks if the content contains a phase blocked marker. This is a convenience function using the default PhaseMarkers matcher.
func IsPhaseComplete ¶ added in v1.1.0
IsPhaseComplete checks if the content contains a phase completion marker. This is a convenience function using the default PhaseMarkers matcher.
Types ¶
type CodeBlock ¶
type CodeBlock struct {
// Language is the language specifier after the opening fence (e.g., "go", "python").
Language string
// Content is the code inside the block, excluding fences.
Content string
// Raw is the complete block including the fences.
Raw string
}
CodeBlock represents a fenced code block.
type Marker ¶ added in v1.1.0
type Marker struct {
// Tag is the marker name (e.g., "phase_complete", "phase_blocked").
Tag string
// Value is the content between the opening and closing tags.
Value string
// Raw is the full matched text including tags.
Raw string
}
Marker represents a detected XML-style marker in content. Markers are used for structured signaling in LLM responses, such as phase completion or blocking indicators.
type MarkerMatcher ¶ added in v1.1.0
type MarkerMatcher struct {
// contains filtered or unexported fields
}
MarkerMatcher finds XML-style markers in content. It compiles regex patterns for each registered tag and caches them for efficient repeated matching.
Example markers:
<phase_complete>true</phase_complete> <phase_blocked>reason: need clarification</phase_blocked> <implement_complete>true</implement_complete>
func NewMarkerMatcher ¶ added in v1.1.0
func NewMarkerMatcher(tags ...string) *MarkerMatcher
NewMarkerMatcher creates a matcher for the given tag names. Tags should be provided without angle brackets.
Example:
matcher := NewMarkerMatcher("phase_complete", "phase_blocked")
func (*MarkerMatcher) AddTag ¶ added in v1.1.0
func (m *MarkerMatcher) AddTag(tag string)
AddTag adds a new tag to match. This is safe for concurrent use.
func (*MarkerMatcher) Contains ¶ added in v1.1.0
func (m *MarkerMatcher) Contains(content, tag string) bool
Contains checks if any marker with the given tag exists in content.
func (*MarkerMatcher) ContainsAny ¶ added in v1.1.0
func (m *MarkerMatcher) ContainsAny(content string) bool
ContainsAny checks if any of the registered markers exist in content.
func (*MarkerMatcher) ContainsValue ¶ added in v1.1.0
func (m *MarkerMatcher) ContainsValue(content, tag, value string) bool
ContainsValue checks if a marker with the specific tag and value exists. The value comparison is case-insensitive and trims whitespace.
func (*MarkerMatcher) FindAll ¶ added in v1.1.0
func (m *MarkerMatcher) FindAll(content string) []Marker
FindAll returns all markers found in content for all registered tags.
func (*MarkerMatcher) FindAllForTag ¶ added in v1.1.0
func (m *MarkerMatcher) FindAllForTag(content, tag string) []Marker
FindAllForTag returns all markers for a specific tag.
func (*MarkerMatcher) FindFirst ¶ added in v1.1.0
func (m *MarkerMatcher) FindFirst(content, tag string) (Marker, bool)
FindFirst returns the first marker found for the given tag. Returns false if no marker is found.
func (*MarkerMatcher) GetValue ¶ added in v1.1.0
func (m *MarkerMatcher) GetValue(content, tag string) string
GetValue extracts the value of a marker with the given tag. Returns empty string if the marker is not found.
func (*MarkerMatcher) Tags ¶ added in v1.1.0
func (m *MarkerMatcher) Tags() []string
Tags returns the list of tags this matcher looks for.
type Parser ¶
type Parser struct {
// contains filtered or unexported fields
}
Parser extracts structured content from LLM responses.
func NewParser ¶
func NewParser() *Parser
NewParser creates a new response parser with compiled regexes.
func (*Parser) ExtractAllCode ¶
ExtractAllCode extracts all code blocks from the response.
func (*Parser) ExtractCode ¶
ExtractCode extracts the first code block with the given language. If language is empty, returns the first code block found.
func (*Parser) ExtractJSON ¶
ExtractJSON extracts and parses the first JSON block found. Returns nil if no valid JSON block is found.
func (*Parser) ExtractJSONArray ¶
ExtractJSONArray extracts and parses JSON arrays from code blocks. Returns all successfully parsed arrays.
func (*Parser) ExtractList ¶
ExtractList extracts list items from the response. Supports both - and * bullet points.
func (*Parser) ExtractNumberedList ¶
ExtractNumberedList extracts numbered list items.
func (*Parser) ExtractSection ¶
ExtractSection extracts the content of a specific section by title.
func (*Parser) ExtractYAML ¶
ExtractYAML extracts and parses YAML blocks.
func (*Parser) HasCodeBlock ¶
HasCodeBlock checks if the response contains any code block.
type Response ¶
type Response struct {
// Raw is the original response text.
Raw string
// Text is the response with code blocks removed.
Text string
// CodeBlocks contains all extracted code blocks.
CodeBlocks []CodeBlock
// JSONBlocks contains parsed JSON blocks.
JSONBlocks []map[string]any
// Sections contains extracted markdown sections by title.
Sections map[string]string
}
Response contains structured data extracted from an LLM response.
Source Files