message

package
v0.5.3 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: May 11, 2026 License: MIT Imports: 3 Imported by: 0

Documentation

Index

Constants

View Source 
const (
	CompactionSummaryHeader = "[Context Summary]\n"
	CompactionCompressedTag = "\n\n[Context compressed]"
	CompactionEvidenceTag   = "[Context Evidence]\n"
	CompactionDisplayHint   = "\n\n[Context display hint]\n"
)

Variables

This section is empty.

Functions

func IsCompactionEvidenceArtifactText 

func IsCompactionEvidenceArtifactText(content string) bool

func IsFileRefContent 

func IsFileRefContent(text string) bool

IsFileRefContent reports whether text is an @-injected <file path="...">...</file> block.

func MergeCompactionSummaryAndEvidence 

func MergeCompactionSummaryAndEvidence(summaryContent, evidenceContent string) string

func UserPromptPlainText 

func UserPromptPlainText(msg Message) string

UserPromptPlainText returns user-visible text for session titles and usage previews: concatenated non–file-ref text parts when Parts is set; otherwise trimmed Content.

Types

type ContentPart 

type ContentPart struct {
	Type        string `json:"type"`                   // "text" or "image"
	Text        string `json:"text,omitempty"`         // for type="text"
	DisplayText string `json:"display_text,omitempty"` // optional TUI-only summary for large hidden text parts
	MimeType    string `json:"mime_type,omitempty"`    // for type="image", e.g. "image/png"
	Data        []byte `json:"data,omitempty"`         // for type="image", raw bytes (not persisted; loaded from ImagePath)
	ImagePath   string `json:"image_path,omitempty"`   // for type="image", path to persisted image file on disk
	FileName    string `json:"file_name,omitempty"`    // optional display name for image attachments
}

ContentPart is one part of a multi-part user message (text or image).

type LSPReview 

type LSPReview struct {
	ServerID string `json:"server_id,omitempty"`
	Errors   int    `json:"errors,omitempty"`
	Warnings int    `json:"warnings,omitempty"`
}

type Message 

type Message struct {
	Role                string             `json:"role"` // "user", "assistant", "tool"
	Content             string             `json:"content"`
	Parts               []ContentPart      `json:"parts,omitempty"`                 // multi-part content (text + images); when set, supersedes Content
	ThinkingBlocks      []ThinkingBlock    `json:"thinking_blocks,omitempty"`       // assistant only; must be replayed verbatim
	ToolCalls           []ToolCall         `json:"tool_calls,omitempty"`            // non-nil for assistant tool_use
	ToolCallID          string             `json:"tool_call_id,omitempty"`          // non-empty for tool results
	ToolDiff            string             `json:"tool_diff,omitempty"`             // unified diff for Write/Edit tool results
	ToolDiffAdded       int                `json:"tool_diff_added,omitempty"`       // total added lines for Write/Edit; computed before diff truncation
	ToolDiffRemoved     int                `json:"tool_diff_removed,omitempty"`     // total removed lines for Write/Edit; computed before diff truncation
	ToolDurationMs      int64              `json:"tool_duration_ms,omitempty"`      // final tool elapsed time in milliseconds for restored footer display
	ToolStatus          string             `json:"tool_status,omitempty"`           // terminal tool status: success|error|cancelled
	FileState           *ToolFileState     `json:"file_state,omitempty"`            // durable file-state metadata for restore-time safety sentinels
	LSPReviews          []LSPReview        `json:"lsp_reviews,omitempty"`           // per-server last-review snapshot for the directly edited file only
	Audit               *ToolArgsAudit     `json:"audit,omitempty"`                 // tool-call audit metadata when effective args differ after confirmation
	IsCompactionSummary bool               `json:"is_compaction_summary,omitempty"` // first user message after compaction (summary of archived history)
	StopReason          string             `json:"stop_reason,omitempty"`           // assistant only; e.g. "stop", "end_turn", "max_tokens", "tool_use"
	Provenance          *MessageProvenance `json:"provenance,omitempty"`            // optional producer/source metadata for model-compat replay decisions
	// Usage is the token usage for this message when it ends an LLM round (assistant only).
	// Persisted in JSONL so session resume can sum per-message usage to restore session totals.
	Usage        *TokenUsage `json:"usage,omitempty"`
	Kind         string      `json:"kind,omitempty"` // control/display subtype, e.g. "loop_notice"
	MailboxAckID string      `json:"-"`              // transient runtime-only mailbox ack marker; never persisted
}

Message represents a conversation message (user, assistant, or tool result).

func RepairOrphanToolResults 

func RepairOrphanToolResults(msgs []Message) ([]Message, int)

RepairOrphanToolResults returns a copy of msgs with tool-role messages removed when no preceding assistant message declared that tool_call_id. It returns the number of dropped messages. Preserves nil vs empty slice: nil input yields nil, 0.

type MessageProvenance added in v0.5.0 

type MessageProvenance struct {
	Source     string `json:"source,omitempty"`      // chord|import:claude|import:codex|import:opencode
	ProviderID string `json:"provider_id,omitempty"` // anthropic-main / openai / gemini ...
	ModelID    string `json:"model_id,omitempty"`
	Variant    string `json:"variant,omitempty"`
	ModelRef   string `json:"model_ref,omitempty"`   // provider/model[@variant]
	WireFamily string `json:"wire_family,omitempty"` // anthropic|openai-chat|openai-responses|gemini|unknown
	Imported   bool   `json:"imported,omitempty"`
}

MessageProvenance captures the producer/source metadata of a persisted message. It is optional and primarily used for request-time model-compat normalization, where provider-specific payloads may need to be replayed, downgraded, or dropped depending on the current target model.

type Response 

type Response struct {
	Content        string
	ThinkingBlocks []ThinkingBlock // non-nil when extended thinking was enabled
	ToolCalls      []ToolCall
	Usage          *TokenUsage
	StopReason     string
	// ThinkingToolcallMarkerHit is true when provider-side reasoning content
	// contained pseudo tool-call template markers (e.g. "<|tool_call_begin|>").
	// This is observational metadata only; tool execution must still come from
	// structured ToolCalls.
	ThinkingToolcallMarkerHit bool
	// ReasoningContent holds the full accumulated reasoning/thinking text from
	// OpenAI-compatible providers. Populated only when ThinkingToolcallMarkerHit
	// is true, so the agent layer can parse pseudo tool calls from it.
	ReasoningContent string
	// ProviderResponseID is the server-assigned response ID from the Responses API
	// (response.completed / response.incomplete). Used by the Codex WebSocket
	// transport for connection-scoped previous_response_id incremental requests.
	ProviderResponseID string
}

Response represents a complete LLM response.

type RollbackDelta 

type RollbackDelta struct {
	Reason string
}

RollbackDelta indicates that the current streamed assistant output should be discarded (e.g. incremental chain validation failed and provider will retry with a full-input request).

type StatusDelta 

type StatusDelta struct {
	Type     string // e.g. "connecting", "waiting_headers", etc.
	Detail   string // e.g. "retry 2/5"
	ModelRef string // non-empty when a model switch is in progress (e.g. fallback: "provider/model")
	Reason   string // optional machine-readable reason for model routing changes (e.g. fallback cause)
}

StatusDelta represents a technical state change during an LLM request.

type StreamDelta 

type StreamDelta struct {
	// Type is one of: "text", "tool_use_start", "tool_use_delta", "tool_use_end",
	// "thinking", "thinking_end", "error", "status", "rate_limits", "rollback",
	// "key_switched", "key_confirmed", "key_deactivated", "key_expired".
	//
	// key_confirmed may carry Status.ModelRef/Status.Reason to identify the
	// effective model that produced the first visible token for the current
	// streaming attempt (used for confirming fallback/key-switch UI toasts).
	Type      string                          // delta category
	Text      string                          // for Type="text" or "thinking"
	ToolCall  *ToolCallDelta                  // for Type="tool_use_*"
	Status    *StatusDelta                    // for Type="status"
	RateLimit *ratelimit.KeyRateLimitSnapshot // for Type="rate_limits"
	Rollback  *RollbackDelta                  // for Type="rollback"
	AccountID string                          // for Type="key_deactivated"/"key_expired": the OAuth account ID
	Email     string                          // for Type="key_deactivated"/"key_expired": the OAuth account email, if available
	Progress  *StreamProgressDelta            // optional cumulative/request progress hint for status bar or transport diagnostics
}

StreamDelta represents an incremental piece of a streaming LLM response.

type StreamProgressDelta 

type StreamProgressDelta struct {
	Bytes  int64 // cumulative response bytes received so far; provider may include response headers
	Events int64 // cumulative stream event/data count received so far
}

StreamProgressDelta reports cumulative response transport progress observed by the provider/parser for the current request.

type ThinkingBlock 

type ThinkingBlock struct {
	Thinking  string `json:"thinking"`
	Signature string `json:"signature"`
}

ThinkingBlock holds extended-thinking content returned by Anthropic models. Both fields must be preserved and sent back verbatim in subsequent API calls so the API can validate the conversation chain.

type TokenUsage 

type TokenUsage struct {
	InputTokens      int `json:"input_tokens"`
	OutputTokens     int `json:"output_tokens"`
	CacheReadTokens  int `json:"cache_read_input_tokens"`
	CacheWriteTokens int `json:"cache_creation_input_tokens"`
	ReasoningTokens  int `json:"reasoning_tokens"`
}

TokenUsage records token usage from a single API response. InputTokens = prompt size (total input); OutputTokens = total generated (content + reasoning if reported together). CacheReadTokens/CacheWriteTokens = cache read/write; ReasoningTokens = thinking/reasoning output when reported separately.

type ToolArgsAudit 

type ToolArgsAudit struct {
	OriginalArgsJSON  string `json:"original_args_json,omitempty"`
	EffectiveArgsJSON string `json:"effective_args_json,omitempty"`
	UserModified      bool   `json:"user_modified,omitempty"`
	EditSummary       string `json:"edit_summary,omitempty"`
}

ToolArgsAudit records how a tool call's effective execution arguments differ from the model's original request after user confirmation.

func (*ToolArgsAudit) Clone 

func (a *ToolArgsAudit) Clone() *ToolArgsAudit

Clone returns a deep copy of the audit payload.

type ToolCall 

type ToolCall struct {
	ID   string          `json:"id"`
	Name string          `json:"name"`
	Args json.RawMessage `json:"args"`
}

ToolCall represents a single tool invocation by the LLM.

type ToolCallDelta 

type ToolCallDelta struct {
	ID    string
	Name  string
	Input string // partial JSON
}

ToolCallDelta represents incremental tool call data during streaming.

type ToolDefinition 

type ToolDefinition struct {
	Name        string         `json:"name"`
	Description string         `json:"description"`
	InputSchema map[string]any `json:"input_schema"`
}

ToolDefinition describes a tool for the LLM API's tools parameter.

type ToolFileState added in v0.5.3 

type ToolFileState struct {
	Reads   []TrackedFileState `json:"reads,omitempty"`
	Writes  []TrackedFileState `json:"writes,omitempty"`
	Deletes []TrackedFileState `json:"deletes,omitempty"`
}

ToolFileState records durable file-state metadata emitted by file tools. It is used only to restore runtime safety sentinels across session restore; file contents are not persisted here or re-injected into model context.

func (*ToolFileState) Clone added in v0.5.3 

func (s *ToolFileState) Clone() *ToolFileState

type TrackedFileState added in v0.5.3 

type TrackedFileState struct {
	Path   string `json:"path"`
	SHA256 string `json:"sha256,omitempty"`
	Exists bool   `json:"exists"`
}

TrackedFileState records the observed state of one file at tool completion.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.