@clear-capabilities/agentic-security-scanner 0.170.0 → 0.172.0

package/CHANGELOG.md

@@ -9,6 +9,44 @@
 > make the history less accurate, not more.
 
 
+## 0.172.0 - Deterministic Context Intelligence (`context` command)
+
+An opt-in, additive subsystem. The default scan, `hunt` and every existing command are unchanged.
+
+- `agentic-security context build` resolves entry points (JavaScript and TypeScript, Python, Haskell, plus the surfaces the existing inventory recognises) and writes one inspectable, hash-bound context bundle each. No model call, no network call. A bundle is `complete` only when nothing material was unresolved, omitted, capped or split.
+- Protection states are explicit per dimension (present, not present, not seen, not resolved, omitted due to budget, evaluated and ineffective for a control that runs after the sensitive operation), with observed versus inferred evidence.
+- `context review` runs focused single-call reviews over frozen bundles through the existing model resolver, egress policy and redaction. Citations are verified against the bundle, severity comes only from the existing confirmation tier, and the bounded queue retries, caches, checkpoints and resumes. A job with no verified result is reported as missing coverage; `--strict` exits 3 on it.
+- `context diff`, `inspect`, `verify`, `export`, `status`, `cleanup`, and an offline batch export and import.
+- Optional read-only compiler metadata import for Haskell (`hie-export/1`), validated against the source snapshot. Automatic metadata builds and a provider batch API are blocked and documented as such.
+- `context eval` scores a bundled synthetic suite with a scripted reader. It is labelled synthetic on its face and supports no accuracy claim; intervals are unmeasured below 10 independent cases, and the holdout split is ledgered.
+- A batch import bug found while testing: a stored declined result no longer blocks importing a good result for the same job.
+- Registered `context-intelligence` and `context-intelligence.json` in the state artifact registry.
+
+See [the guide](docs/guides/context-intelligence.md).
+
+## 0.171.0 - documentation: surface every shipped PRD capability in the README and docs index
+
+A documentation-only release. Auditing the docs against the actual CLI surface after PRD 5 shipped found
+several real gaps — not cosmetic, actively misleading in one case — and fixed them:
+
+- **README.md said built container images are NOT scanned.** That line predated `image scan`/`image verify`
+  (PRD 5, shipped in 0.170.0) and was flatly false by the time it shipped. Corrected, with a link to the new
+  guide below.
+- **Seven of the repo's 17 slash-command dispatchers had no entry in README's own Commands list**:
+  `dataflow`, `governance`, `remediation`, `federate`, `image`, `provenance`, `runtime`. Added, grouped under
+  two new categories ("Data Flow Explorer" and "Source-to-artifact & runtime").
+- **`docs/guides/runtime-twin.md` (PRD 2's own user guide) was an orphan** — reachable from one internal
+  cross-link, linked from neither README.md nor docs/README.md. Linked from both.
+- **PRD 5 (Source-to-Artifact Security & Provenance) had no user-facing guide at all**, only the internal
+  `docs/implementation-status/` tracking doc. Added `docs/guides/artifact-provenance.md`.
+- **`docs/reference/cli.md` — "every command, flag, and exit code" — was missing `proof`, `rules`, `fleet`,
+  `runtime`, `image`, `provenance`, and `supply` entirely** from its own Commands table (the literal CLI
+  surface of PRDs 1, 3, 4, 2, and 5). Added all seven.
+- Added two new "What makes it different" bullets (runtime corroboration, source-to-artifact tracing) and a
+  `supply --vex` mode mention to README's Commands list.
+
+No code changed; `scanner/dist/agentic-security.mjs` is unchanged.
+
 ## 0.170.0 - source-to-artifact security & provenance: a new `src/artifact/` package, `image`/`provenance` CLI, VEX
 
 Binds a finding to the actual **built artifact** (a container image, today) that ships it — closing the gap

package/bin/agentic-security.js

@@ -181,6 +181,11 @@ Commands:
                                to each finding of a scan result. Read-only, local files only.
                                Off by default: set AGENTIC_SECURITY_ASSURANCE_DEPLOYMENT_BOUNDARIES=1.
                                Exit: 0 ok, 1 failed, 2 usage, 3 feature off or blocked.
+  context <build|inspect|diff|review|status|eval|verify|export|cleanup> [--root <dir>]
+                               Deterministic Context Intelligence (opt-in, additive): resolve entry points,
+                               write one inspectable, hash-bound context bundle each, and optionally run
+                               focused single-call model reviews over them. 'context build' makes no model
+                               call and no network call. Run 'agentic-security context' for the full usage.
   explore [path] [--port <n>] [--keep-open]
                                Start a local, read-only server over an
                                already-scanned lineage graph (run
@@ -7636,6 +7641,7 @@ async function main() {
       case 'calibration-feedback': process.exit(await cmdCalibrationFeedback(args));
       case 'calibration-report': process.exit(await cmdCalibrationReport(args));
       case 'hunt':     process.exit(await cmdHunt(args));
+      case 'context':  process.exit(await (await import('../src/context-intelligence/cli.js')).cmdContext(args));
       case 'compliance': process.exit(await cmdCompliance(args));
       case 'attest':   process.exit(await cmdAttest(args));
       case 'verify-attestation': process.exit(await cmdVerifyAttestation(args));