pub struct WalkthroughGuide<Digest> {
pub schema_version: ReviewBriefSchemaVersion,
pub version: String,
pub command: String,
pub graph_snapshot_hash: String,
pub digest: Digest,
pub direction: ReviewDirection,
pub change_anchors: Vec<ChangeAnchor>,
pub agent_schema: AgentSchema,
pub injection_note: &'static str,
}Expand description
The fallow review --walkthrough-guide envelope: the current digest + schema
the agent fetches. The tool owns this; the skill stays thin (it fetches this
rather than embedding a frozen copy). Always emitted with exit 0.
Fields§
§schema_version: ReviewBriefSchemaVersionPinned to the brief schema version (the spec versions the guide by
review_brief_schema_version).
version: StringFallow CLI version that produced this guide.
command: StringCommand discriminator singleton: always "review-walkthrough-guide".
graph_snapshot_hash: StringThe deterministic graph-snapshot hash pinned into the digest. The agent echoes it back; a mismatch on reentry refuses the payload as stale.
digest: DigestThe graph-derived digest (brief + decision surface). Pure over the tree.
direction: ReviewDirectionThe review direction (order/units/concern-lens/out-of-diff/expert).
change_anchors: Vec<ChangeAnchor>The per-hunk change anchors: one stable id per changed region. An agent
may cite a change_anchor as a judgment anchor in addition to an emitted
signal_id, so a trade-off about a changed region with no graph finding
can still anchor (and be post-validated) rather than hallucinate.
agent_schema: AgentSchemaThe JSON shape the agent must return, embedded so the skill stays thin.
injection_note: &'static strThe injection-resistance note (digest is graph-only; PR prose untrusted).