Struct PromptBundle 

pub struct PromptBundle {
    pub fragments: Vec<PromptFragment>,
    pub environment: EnvironmentMetadata,
    pub project_context: Vec<ContextSource>,
    pub tool_catalog: Vec<ToolDefinition>,
    pub available_skills: Vec<SkillMetadata>,
    pub transcript_tail: Vec<Entry>,
    pub user_turn: String,
    pub history_reuse: HistoryReuse,
    pub prev_context_hash: Option<u64>,
    pub context_ledger: Option<ContextLedger>,
}

The structured prompt bundle before provider-specific lowering.

Each field is a separate piece of model context. The std::fmt::Display implementation renders the full system prompt text; lower_to_umans_messages converts it to Anthropic-compatible messages.

Fields

fragments: Vec<PromptFragment>

Ordered prompt fragments: base identity, communication style, action model, edit guidance, action safety, self-knowledge, web/source guidance.

environment: EnvironmentMetadata

Environment: cwd, model, search mode, rounded date.

project_context: Vec<ContextSource>

Loaded AGENTS.md context sources.

tool_catalog: Vec<ToolDefinition>

Tool catalog as provider-native schemas.

available_skills: Vec<SkillMetadata>

Available Agent Skills metadata. Full skill instructions are read only after activation.

transcript_tail: Vec<Entry>

Projected model-visible transcript tail.

user_turn: String

Current user prompt text.

history_reuse: HistoryReuse

Whether the provider supports reusable history / prompt caching for AGENTS.md content. When HistoryReuse::Available, full text is included only when the content hash changes.

prev_context_hash: Option<u64>

Content hash of the root AGENTS.md from the previous turn, if any.

Used with HistoryReuse::Available to skip re-sending unchanged AGENTS.md text. None on the first turn or when no context is loaded.

context_ledger: Option<ContextLedger>

Selected context projection from the context selection policy (P7/P8).

When Some, the system prompt renders selected project instructions, compaction summaries, and pinned handles from the ledger items, the transcript tail is taken from ledger-selected transcript context, and a compact context dashboard is added to <thndrs_self_knowledge>.

When None, the legacy rendering path is used (project_context plus the fixed transcript tail projection), preserving existing behavior.

Implementations

impl PromptBundle

pub fn new( cwd: &Path, model: &str, mode: WebSearchMode, context_sources: &[ContextSource], transcript: &[Entry], user_turn: &str, ) -> PromptBundle

Build a PromptBundle from app state and context.

pub fn new_with_skills( cwd: &Path, model: &str, mode: WebSearchMode, context_sources: &[ContextSource], available_skills: &[SkillMetadata], transcript: &[Entry], user_turn: &str, ) -> PromptBundle

Build a PromptBundle including discovered Agent Skills metadata.

pub fn with_tool_catalog(self, tool_catalog: Vec<ToolDefinition>) -> Self

Replace the bundle’s tool catalog with a runtime-specific catalog.

pub fn with_context_ledger(self, ledger: ContextLedger) -> Self

Attach a selected context projection (P7 ledger) so the system prompt renders selected instructions, summaries, pins, and transcript context from the ledger, and adds a context dashboard to <thndrs_self_knowledge>.

Trait Implementations

impl Clone for PromptBundle

fn clone(&self) -> PromptBundle

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for PromptBundle

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl From<&PromptBundle> for SelfKnowledgeSnapshot

fn from(bundle: &PromptBundle) -> SelfKnowledgeSnapshot

Converts to this type from the input type.