Module codewiki

Structs§

CitationRepairSummary: Result of a citation-repair run. This is the source-of-truth serialized shape that the gcode contract freezes for codewiki --repair-citations (Leaf 5 / #876): a stable key set with no dry_run mode.
CodewikiAiOptions
CodewikiGraphEdge
CodewikiInput
CodewikiRunSummary
Crate: A workspace member crate.
Edge: A workspace-internal dependency edge: from depends on to, where both endpoints are member package names. Edges to external crates.io dependencies are never recorded.
LeadingChunk: The first indexed content chunk of a file: real source text with its line range, used as retrieved prompt input and citation provenance.
ServiceBoundary: One service boundary the workspace reaches, plus what pulls it in.
SystemModel: Deterministic, serializable model of the workspace. Built with no LLM calls from facts on disk.

AiDepth: How deep AI prose generation reaches. Deeper tiers include shallower ones; gated tiers fall back to structural summaries.
CodewikiGraphAvailability
CodewikiGraphEdgeKind
PromptTier: Weight tier of one codewiki generation call (#904). Aggregate is the top-level repo-wide synthesis — repo overview, architecture, and the curated narrative/concept layer — and is written opus-first. Module is mid-level per-unit synthesis (module docs and file-body narratives) and routes to sonnet. Standard is high-volume per-symbol prose on the default low tier.
ProseDepth: Output verbosity for AI prose, orthogonal to AiDepth (which page tiers reach the LLM) and to the audience register. Maps to a per-page output token budget; ProseDepth::Standard defers to the provider/profile default so a run without the flag is byte-identical to before this control existed.
ProseRegister: Audience register for AI prose, orthogonal to depth. Every register projects the same grounded facts and only changes voice; None (the default) leaves the base system prompts untouched so default runs are unchanged.
RuntimeMode: How a binary can run with respect to the AI routing decision.
ServiceKind: A runtime service / external boundary the workspace can talk to.

build_system_model: Build a SystemModel from the workspace rooted at repo_root.
generate_hierarchical_docs
repair_citations: Public entry: re-anchors every generated page’s citations against the current symbols, using the index snapshot persisted in the vault meta to identify which symbol each stale citation named. No regeneration, no LLM — this is the routine the codewiki --repair-citations flag (Leaf 5 / #876) drives. A vault with no persisted snapshot (unwrap_or_default) cannot identify moved symbols, so its stale citations all count as unresolved.
run
run_repair: Repair-only entry for codewiki --repair-citations: re-anchors every generated page’s [file:line] citations against the current index and rewrites only the pages whose citations changed. No generation, no AI/LLM calls. Loads the full visible symbol set (like run) so a citation to any indexed file can resolve, then prints the super::CitationRepairSummary.
write_doc_set
write_incremental_doc_set

TextGenerator
TextVerifier: Grounded verification call: given a verify prompt and system prompt, returns the raw model response, or None when the verifier is unavailable (routed off, transport failure, or generation error). Callers treat None as “skip verification, proceed undegraded”. The deterministic block numbering, response parsing, and stripping live in [super::text], so the closure is just the model call — mirroring TextGenerator but without a prompt tier.