Documentation
¶
Overview ¶
Package manifest is the on-disk index metadata. CKV writes <out>/manifest.json alongside <out>/vector.db so freshness checks, rebuild detection, and CKG cross-checks can run without opening SQLite.
Schema versioning: bump SchemaVersion only on *breaking* changes (renamed fields, removed fields, changed semantics). Additive `omitempty` fields stay on the same version — old readers see them as zero values.
Index ¶
Constants ¶
const FileName = "manifest.json"
FileName is the file written into the --out directory.
const SchemaVersionCurrent = "1.0"
SchemaVersionCurrent is the on-disk schema version this build writes.
Variables ¶
var ErrNotFound = errors.New("manifest: not found")
ErrNotFound signals that no manifest exists at the given path.
Functions ¶
func Remove ¶
Remove deletes <dir>/manifest.json if present. A missing file is not an error, so callers can use it to mark an index "not ready" before a rebuild: while the build runs (or if it fails partway) Load returns ErrNotFound, keeping a partially-written index from being opened against a stale manifest.
Types ¶
type CKGSource ¶
type CKGSource struct {
GraphDigest string `json:"graph_digest,omitempty"`
SrcCommit string `json:"src_commit,omitempty"`
SchemaVersion string `json:"schema_version,omitempty"`
Path string `json:"path,omitempty"`
}
CKGSource anchors the CKG graph this index aligned against, for CKG↔CKV mismatch detection (design §3). GraphDigest is CKG's *logical* digest (sorted canonical_id + edge hash — NOT a file-byte sha, which is non-deterministic under SQLite's layout). It stays empty until CKG publishes it; SrcCommit + SchemaVersion are recorded immediately, so the alignment assert starts on commit+schema and strengthens to +digest once available.
type CodeSource ¶
type CodeSource struct {
IndexedHead string `json:"indexed_head,omitempty"`
BuiltAt string `json:"built_at,omitempty"`
}
CodeSource is the source-tree cutoff (mirrors the top-level SrcCommit/BuiltAt, grouped for a uniform ledger).
type HashedSource ¶
type HashedSource struct {
Path string `json:"path,omitempty"`
ContentHash string `json:"content_hash,omitempty"` // sha256 hex
}
HashedSource is a corpus/config whose "did it change" is detected by a content hash (docs tree, flow corpus file, policy file).
type Manifest ¶
type Manifest struct {
SchemaVersion string `json:"schema_version"`
CKVVersion string `json:"ckv_version"`
BuiltAt string `json:"built_at"` // RFC3339
// Source / git
SrcRoot string `json:"src_root"`
SrcCommit string `json:"src_commit,omitempty"` // commit at indexing time
IndexedHead string `json:"indexed_head,omitempty"` // alias for SrcCommit (back-compat)
// Embedding identity
EmbeddingModel string `json:"embedding_model"`
EmbeddingDim int `json:"embedding_dim"`
EmbeddingChecksum string `json:"embedding_checksum,omitempty"`
EmbeddingNormalize string `json:"embedding_normalize,omitempty"` // "l2" | "none"
// Aggregate stats
ChunkCount int `json:"chunk_count"`
Languages map[string]int `json:"languages,omitempty"` // language → chunk count
// Ignore patterns surfaced for transparency
CKVIgnore []string `json:"ckvignore,omitempty"`
// DocsRoots are additional markdown corpus directories indexed via
// `ckv build --docs` (outside SrcRoot, e.g. a cks-rendered
// domain-knowledge corpus). Recorded so callers can see every source
// the index covers. Additive — old readers see nil.
DocsRoots []string `json:"docs_roots,omitempty"`
// Sources is the per-layer knowledge-cutoff ledger (reindex-migration
// design §2.2): what each layer was built from, so a reindex knows what
// is stale and CKS can detect a CKG↔CKV mismatch. Additive — old readers
// see nil; each sub-block is omitted when that layer was not built.
Sources *Sources `json:"sources,omitempty"`
}
Manifest is the structured index metadata. Field names are shared with CKG so CKS Orchestrator can compare CKG and CKV manifests by raw key.
type PRSource ¶
type PRSource struct {
Repo string `json:"repo,omitempty"`
LastPRNumber int `json:"last_pr_number,omitempty"`
LastMergedAt string `json:"last_merged_at,omitempty"` // RFC3339
}
PRSource is the PR-corpus cutoff: the newest PR indexed, so an incremental PR ingest can fetch only PRs after LastPRNumber / LastMergedAt.
type Sources ¶
type Sources struct {
Code *CodeSource `json:"code,omitempty"`
CKG *CKGSource `json:"ckg,omitempty"`
PRs *PRSource `json:"prs,omitempty"`
Docs *HashedSource `json:"docs,omitempty"`
Flow *HashedSource `json:"flow,omitempty"`
Policy *HashedSource `json:"policy,omitempty"`
}
Sources records the origin + cutoff of each knowledge layer in the index.
Source Files