Skip to content

OpenAI's Codex Security CLI and TypeScript SDK for finding, validating, and fixing security vulnerabilities. npm: https://www.npmjs.com/package/@openai/codex-security

release: bump Codex Security to 0.1.25 (#751)

mldangelo-oai7d042cdaefdb
diff --git a/.github/release-notes.md b/.github/release-notes.md
index a0af0767d..16dd0ca8d 100644
--- a/.github/release-notes.md
+++ b/.github/release-notes.md
@@ -1,43 +1,74 @@
-<!-- release-version: 0.1.24 -->
+<!-- release-version: 0.1.25 -->
## Highlights
-- Start the preview findings service directly with
- `codex-security serve [--port PORT]`, without Docker or an internal package
- path. The command reuses the existing service, state, and shutdown behavior;
- `--port` overrides `PORT`, and port `0` selects a free port. See
- [running without Docker](https://github.com/openai/codex-security/blob/npm-v0.1.24/sdk/typescript/README.md#running-without-docker).
-- Observe durable Deep Scan progress from the SDK with the optional
- `onDeepProgress({ completed, active, maximum })` callback. Updates report
- changed completed and active independent-review counts without blocking the
- scan. See
- [SDK scan options](https://github.com/openai/codex-security/blob/npm-v0.1.24/sdk/typescript/README.md#sdk-configuration-and-scan-options).
-- Make stopped-result handling explicit and stable. Read, list, and export
- operations no longer publish late retained results as a side effect; the app
- reports when recovery is needed and can recover validated results on request.
- See
- [stopped result recovery](https://github.com/openai/codex-security/blob/npm-v0.1.24/plugins/codex-security/references/scan-contract.md#stopped-result-recovery).
-- Include changed PowerShell `.ps1` files in diff-scan inventories and remove a
- conflicting reporting rule so valid internal attack paths remain eligible for
- review. Nested Deep Scan workers now also receive an explicitly configured
- OpenAI provider credential through the plugin's existing environment boundary.
-- Improve Windows reliability by preserving case-insensitive `CODEX_HOME`
- entries and repository paths from ordinary PowerShell activity, and by
- retrying credential snapshots when a descendant file disappears during ACL
- inspection. Existing path-safety and permission failures remain fatal.
+- Preserve confirmed finding identities across scans and comparisons, and show
+ related findings with their reasons while keeping distinct findings separate.
+ Large comparisons now use bounded batches without truncating finding text;
+ inputs that cannot fit leave matching explicitly incomplete.
+- Improve deduplication with separate screening and pair reviews, validated
+ pair assignments, and groups that respect explicit `DISTINCT` decisions.
+ Invalid submissions receive one corrective turn; blocked reviews fail without
+ recording a verdict. `DeduplicationReviewError` exposes structured, sanitized
+ failure details. The SDK also adds `deduplicateScanDirectory` for complete,
+ sealed scans outside local history.
+- Generate synthetic Standard scan results with `scan --mock` or the SDK's
+ `mock: true`, without authentication or model calls. Mock results support
+ normal reports, exports, history, and reruns. See
+ [mock scans](https://github.com/openai/codex-security/blob/npm-v0.1.25/sdk/typescript/README.md#generate-mock-scan-results).
+- Increase a running scan's total budget from the interactive dashboard when
+ usage reaches 80% of its limit, or use the SDK's `onBudgetApproaching`
+ callback. The existing limit remains enforced until an increase is saved.
+ See [scan cost limits](https://github.com/openai/codex-security/blob/npm-v0.1.25/sdk/typescript/README.md#progress-and-cost).
+- Recognize existing Codex authentication in CLI and SDK login status. SDK
+ scans, comparisons, and deduplication reviews now honor native command-auth
+ providers, including renewable tokens.
+- Configure the findings service's full embeddings endpoint with
+ `CODEX_SECURITY_EMBEDDINGS_URL`. The new `@openai/codex-security/server`
+ exports support embedding credentials supplied by a callback before each
+ HTTP batch. See
+ [embeddings and storage](https://github.com/openai/codex-security/blob/npm-v0.1.25/sdk/typescript/README.md#embeddings-and-storage).
+- Include PowerShell module (`.psm1`) and data (`.psd1`) files in scan
+ inventories, and recognize BOM-marked UTF-16 source files as text.
+- Preserve scoped scan and component-plan inventories after directory renames
+ that change only letter casing on case-insensitive filesystems.
+- Support long Codex executable paths on Windows, including nested Deep Scan
+ workers, and retry credential snapshots for another `Get-Acl` path-not-found
+ race when a descendant disappears during inspection.
+- Honor case-insensitive Windows environment variable names during finding
+ deduplication, so configured API credentials and private configuration paths
+ are used consistently.
+- Stream tracked binary diffs when hashing repository snapshots, reducing
+ memory use while preserving the existing digest format.
+- Include complete OCI metadata in container image labels and multiarchitecture
+ annotations, with documentation pinned to the source commit and image
+ verification commands in the release workflow summary. See
+ [container metadata and verification](https://github.com/openai/codex-security/blob/npm-v0.1.25/docker/README.md#image-metadata-and-verification).
## Upgrade notes
-- The findings API and dashboard still have no built-in authentication.
- `codex-security serve` binds to loopback by default; keep it local or place it
- behind an authenticated TLS proxy before sharing access. Python is still
- required, and nonempty imports still require an embeddings API credential.
-- Stopped-scan recovery is now explicit. App clients should check
- `resultsRecoveryNeeded` and request recovery when they want validated late
- results republished. Canceled scans remain immutable and cannot use this
- recovery path.
-- `onDeepProgress.maximum` is the configured independent-review cap, not a
- percentage denominator. The SDK polls the durable projection only when the
- callback is supplied.
+- Mock mode is opt-in and saves clearly marked synthetic findings in local
+ history; use a separate `CODEX_SECURITY_STATE_DIR` for disposable test data.
+ It supports Standard scans only and does not audit the repository.
+- Interactive budget increases are unavailable in CI, JSON/JSONL, headless,
+ and verbose modes. Existing cost limits continue to apply in those modes.
+- The embeddings URL defaults to the existing OpenAI endpoint. A configured
+ endpoint receives finding inputs and the bearer credential and must support
+ the OpenAI embeddings format. Embeddings credentials remain separate from
+ Codex ChatGPT sign-in.
+- Local history applies an automatic database index migration. Completed scan
+ artifacts remain unchanged.
+- Source builds now use repository-pinned pnpm 11.19.0, including MCP app
+ dependencies, whose configuration requires a seven-day minimum release age.
+ From the repository root, run
+ `pnpm --dir plugins/codex-security/mcp-app install --frozen-lockfile`.
+ See
+ [running without Docker](https://github.com/openai/codex-security/blob/npm-v0.1.25/sdk/typescript/README.md#running-without-docker).
+- Container publication remains separate from npm publication. Existing stable
+ container tags are not updated in place.
+
+Build and CI updates also improve package verification, portable Python checks,
+Windows fixtures, and test scheduling. Documentation clarifies portable
+environment-variable guidance and safe examples.
The categorized list below contains the individual changes.
diff --git a/sdk/typescript/package.json b/sdk/typescript/package.json
index c6f477c03..721d88495 100644
--- a/sdk/typescript/package.json
+++ b/sdk/typescript/package.json
@@ -1,6 +1,6 @@
{
"name": "@openai/codex-security",
- "version": "0.1.24",
+ "version": "0.1.25",
"description": "TypeScript SDK and CLI for Codex Security",
"license": "Apache-2.0",
"author": "OpenAI",