data/ is deepsec's on-disk state. Each project owns a
subdirectory. The format is wire-compatible with the original
TypeScript implementation — same field casing, same nesting — so a
data/ directory written by either is readable by both.
data/<projectId>/
├── project.json
├── tech.json # detected framework / language tags
├── files/<relative/path>.json # one FileRecord per source file
├── runs/<runId>.json # one RunMeta per scan/process/revalidate
└── reports/
├── report.md
├── report.json
└── report.csv
Append-only per-file accumulator. Fields are camelCase on disk:
| Field | Meaning |
|---|---|
filePath |
Project-relative path (forward slashes). |
projectId |
Owning project id. |
candidates[] |
Scanner-emitted regex hits: vulnSlug, lineNumbers[], snippet, matchedPattern. |
lastScannedAt |
ISO timestamp of the last scan that touched this file. |
lastScannedRunId |
RunId of that scan. |
fileHash |
SHA-256 hex of file contents. |
findings[] |
Real findings produced by the AI processor. |
analysisHistory[] |
Append-only log of investigations (cost, tokens, agent, refusal). |
gitInfo |
Recent committers + ownership (optional, populated by enrich). |
status |
pending / processing / analyzed / error. |
lockedByRunId, lockedAt |
Set when a process run claims the file. |
One per CLI invocation that does write-work.
| Field | Meaning |
|---|---|
runId |
YYYYMMDDHHMMSS-<4hex>. |
type |
scan / process / revalidate. |
phase |
running / done / error. |
scannerConfig |
Set for scan: which matchers ran, mode (full vs files). |
processorConfig |
Set for process / revalidate: agent type, model, invocation mode, source label. |
stats |
filesScanned, candidatesFound, totalCostUsd, totalInputTokens, totalOutputTokens, totalDurationMs, … |
Commit data/ so PR-time scans can compare against the baseline.
deepsec data-commit runs git add data/ && git commit for you with a
default message.