Skip to content

armgit / AiSOC

Public Imported from GitHub
0

Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.

chore(release): v13.0.1 (#1063)

Beenu Arorab3b7a8c367a0
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 84db651eb..fafcfa4f9 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,8 +7,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
## [Unreleased]
+## [13.0.1] - 2026-09-29
+
### Fixed
+- **The corrected chart existed only in the tree: `6.0.0` was never published, and the only chart anybody could pull was the overwritten `5.9.2`.** The fix for the v13.0.0 overwrite landed on `main` *after* the v13.0.0 tag, and the chart reaches GHCR from exactly one place — the `chart-publish` job in `release.yml`, whose `Log in to GHCR`, `Push to the OCI registry` and `Prove it resolves` steps are each gated `if: github.event_name == 'push'`. A `workflow_dispatch` against the existing tag therefore packages, lints, checks the chart against the registry and stops, printing `Chart … was packaged and linted but not pushed`. That is the design working — a dispatch must not republish a version the registry already holds — but it means a fix that lands after a tag has no way to the registry except a new tag. Measured before this release: `oci://ghcr.io/beenuar/charts/aisoc` held `5.6.0`, `5.7.0`, `5.8.0`, `5.9.0`, `5.9.1`, `5.9.2` and nothing else, while the tree read `6.0.0`. This release is the tag push that publishes it.
+
+ **The chart is `6.0.1`, not `6.0.0`.** The 5.x → 6.x boundary is what announces the v13.0.0 behavioural break to anyone upgrading off the published `5.9.2`, and that boundary is crossed whichever 6.x is the first one pullable — semver's contract is the major, not which patch within it happens to publish first. From `6.0.0` the increment is a *patch* because that is what moved inside `infra/helm/aisoc/`: `appVersion` naming a patch release, and nothing else. No template changed, no default changed, no `values.yaml` key was added, removed or renamed, so no consumer's values file breaks — which rules out a major — and no chart capability was added, which rules out a minor. `6.0.0` is deliberately not reused even though the registry never held it: it is already declared on `main` against `appVersion: v13.0.0`, and `scripts/check_chart_version.py` reads git rather than the registry precisely so that an unpublished-but-declared version still counts as taken. Unpublished is not the same as free, and a version naming two applications is the defect the `5.9.2` overwrite taught.
+
- **Nothing read the Helm chart's own `version`, so `scripts/check_chart_version.py` now does.** The entry below records the incident; this records the gate, because a fix that bumps one number leaves the next release free to make the same mistake. The property enforced is not "the two numbers moved together" — that is a proxy, and a proxy passes whenever somebody bumps the wrong one — but **a chart version is never reused for different content**. Four checks read git: `appVersion` moved while `version` did not; a tracked chart file moved with no version bump (Chart.yaml's own `version:` is masked out of that comparison, so the bump cannot be the change that justifies itself); a version that moved backward; and one chart version declared at two release tags with differing content. A fifth asks GHCR whether the version about to be pushed already exists holding something else, and it runs in `release.yml` **before** `helm push`, with `--require-network`, because the push overwrites without a word and there is no after.
Comparing a published chart to a tree needed two structural facts handled rather than assumed. `helm package` is not byte-reproducible, so comparing layer digests would report a difference on a chart nobody touched; the comparison is on unpacked files. `helm package` also re-serialises `Chart.yaml`, dropping comments, so that one file is compared parsed and every other byte for byte — and the tracked subchart archives are expanded the way helm expands them. Against the published 5.9.2 that mapping is exact: 146 files, and the only difference is the one being looked for.
@@ -19,7 +25,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
Fixed by a lockfile bump to 7.4.6 rather than a rebuild: `[email protected]` declares `^7.4.2`, which already permitted the patched release. A `webpack-dev-middleware: ">=7.4.6 <8"` floor joins the root `pnpm.overrides` beside the `webpack-dev-server` entry so a future re-resolution cannot walk back, which a bare lockfile edit would not prevent. One package moved in the lockfile and nothing else; `pnpm install --frozen-lockfile` was verified locally before the push. Measured with the pinned Trivy 0.74.0: 1 HIGH before, 0 after.
-- **The Helm chart's own `version` did not move with v13.0.0, so `helm push` republished an existing chart version pointing at a different application.** Every release before it bumped both coordinates — `appVersion` to the application tag and `version` up the chart's own ladder (5.9.0 → 5.9.1 → 5.9.2) — and v13.0.0 bumped only `appVersion`, so `ghcr.io/beenuar/charts/aisoc:5.9.2` now means `v12.3.2` or `v13.0.0` depending on when it was pulled. A chart version is immutable by convention and nothing in CI enforces that it moves, which is why this was invisible: `scripts/check_published_images.py` reads `appVersion` because that is what becomes an image tag, and never looks at `version`. The chart is now **6.0.0** — a major rather than a minor, because the application it installs has a breaking change and a chart consumer running `helm upgrade` gets it whatever the templates did. The already-overwritten 5.9.2 cannot be un-published; this makes the coordinate honest from here on.
+- **The Helm chart's own `version` did not move with v13.0.0, so `helm push` republished an existing chart version pointing at a different application.** Every release before it bumped both coordinates — `appVersion` to the application tag and `version` up the chart's own ladder (5.9.0 → 5.9.1 → 5.9.2) — and v13.0.0 bumped only `appVersion`, so `ghcr.io/beenuar/charts/aisoc:5.9.2` now means `v12.3.2` or `v13.0.0` depending on when it was pulled. A chart version is immutable by convention and nothing in CI enforces that it moves, which is why this was invisible: `scripts/check_published_images.py` reads `appVersion` because that is what becomes an image tag, and never looks at `version`. The chart moved to **6.x** — a major rather than a minor, because the application it installs has a breaking change and a chart consumer running `helm upgrade` gets it whatever the templates did. That fix set `6.0.0`, which never reached the registry because it landed after the tag; the first published 6.x is the `6.0.1` described at the top of this section, and the major boundary it announces is the same one. The already-overwritten 5.9.2 cannot be un-published; this makes the coordinate honest from here on.
## [13.0.0] - 2026-09-29
diff --git a/README.md b/README.md
index 1d4f2f1cc..358f0c979 100644
--- a/README.md
+++ b/README.md
@@ -7,7 +7,7 @@
**An open-source, self-hostable AI Security Operations Center.** It ingests your security telemetry, detects and correlates threats, investigates them with AI agents whose reasoning is fully auditable, and proposes responses a human approves.
[![License: MIT](https://img.shields.io/badge/License-MIT-22c55e.svg?style=flat-square)](https://opensource.org/licenses/MIT)
-[![Version](https://img.shields.io/badge/version-13.0.0-f59e0b?style=flat-square)](CHANGELOG.md)
+[![Version](https://img.shields.io/badge/version-13.0.1-f59e0b?style=flat-square)](CHANGELOG.md)
[![CI](https://img.shields.io/github/actions/workflow/status/beenuar/AiSOC/ci.yml?branch=main&label=CI&style=flat-square)](https://github.com/beenuar/AiSOC/actions/workflows/ci.yml)
[![CodeQL](https://img.shields.io/github/actions/workflow/status/beenuar/AiSOC/codeql.yml?branch=main&label=CodeQL&style=flat-square)](https://github.com/beenuar/AiSOC/actions/workflows/codeql.yml)
[![OpenSSF Scorecard](https://api.securityscorecards.dev/projects/github.com/beenuar/AiSOC/badge)](https://securityscorecards.dev/viewer/?uri=github.com/beenuar/AiSOC)
diff --git a/RELEASES.md b/RELEASES.md
index 99cee843b..b486ec941 100644
--- a/RELEASES.md
+++ b/RELEASES.md
@@ -2,13 +2,26 @@
This file mirrors what used to live in the "What's new" section of [`README.md`](README.md). The complete, machine-readable inventory (with file paths, env-var diffs, and per-release test counts) lives in [`CHANGELOG.md`](CHANGELOG.md).
-> **TL;DR for first-time visitors:** AiSOC is on `v13.0.0`, released 2026-09-29. **It is a major because it breaks behaviour, not a schema:** 75 state-changing API routes now require a permission they did not require before, and unlike the two patches earlier the same day, some of them refuse roles other than `viewer` — several surfaces move to `settings:write`, which `soc_analyst`, `soc_lead` and `threat_hunter` do not hold, so operators who are not tenant administrators will meet HTTP 403 where they previously got 200. The `### BREAKING` section of [`CHANGELOG.md`](CHANGELOG.md) lists every affected route, the permission it now takes, the roles that lose it, and what to do; `docs/openapi.yaml` is unchanged apart from one corrected description, which is exactly why the break is announced rather than left to be inferred from the spec. What it fixes is the debt the previous two patches measured but did not work down: `scripts/check_route_authz.py` reported **103 of 246 state-changing routes in `services/api` authenticating the caller and then checking nothing** — 42% of the write surface — and that figure is now **28**, every remaining route carrying a recorded reason. The worst module was MSSP, where `_require_own_child` established whose child a tenant was and nothing established whether the caller could act on it, so a `viewer` in a managing tenant could push a rule pack into a customer, grant itself a role inside one, or delete a critical detection from one. `community.py` had the moderation step gated and the thing being moderated open. Every permission used already existed and was already withheld from `viewer`, so there is no migration and no new permission to define. Upgrading is `make up`. v12.3.2 earlier the same day was the second security patch of the day: eleven more write routes were found declaring a permission that never ran. `Annotated[Any, require_permission("users:write")]` carries no `Depends()`, and FastAPI honours only `Annotated` metadata that is a `Depends` or a `FieldInfo` — everything else is silently dropped, so the check never happened and `current_user` degraded into a required query parameter. The tell was the status code: a `viewer` hitting these routes got **422, not 403**, which reads as a malformed request rather than a refusal. v12.3.1 was a security patch on top of v12.3.0, and the reason it exists separately is that a fix living on `main` is not a fix anyone has. Two reported advisories: a **critical** SQL injection in the osquery allowlist, where a denylist scanning rendered SQL for `--`, `/*` and `;` had never listed `'`, so a value could close a quoted literal and replace the WHERE clause — and reproducing it showed all six query templates were injectable, not the one reported, because the numeric parameters were never coerced either. Each parameter now declares what it is and anything else is refused rather than escaped, because an escaped path still reaches the endpoint, matches nothing, and reads to an operator as a clean host. The second is a **high** missing-authorization hole where every remediation write route authenticated and none authorized, so a `viewer` could raise the tenant to autonomy L4 and pre-approve a high blast-radius verb; the permission it needed already existed and was simply never enforced, which is the third time on this repository. The gate that should have caught it passed throughout, because it asks whether a route authenticates — so a second gate now asks whether it authorizes, and measures 103 of 246 state-changing routes making no authorization decision, as a ceiling that can only fall. Upgrading is `make up`. v12.3.0 (same day) closed six lettered sub-phases; its notes are below. None of the eight first-party npm/PyPI packages has been uploaded — the registries return 404 for all eight — and that remains credential-blocked rather than done. The latest GitHub release with notes and downloads: <https://github.com/beenuar/AiSOC/releases/latest>.
+> **TL;DR for first-time visitors:** AiSOC is on `v13.0.1`, released 2026-09-29. It is a patch that exists to **publish the corrected Helm chart**, because until it the only chart anybody could pull was the wrong one. v13.0.0 bumped the chart's `appVersion` and left its own `version` at `5.9.2`, so `helm push` — which replaces silently — republished a coordinate v12.3.2 already held, pointing at a different application. The fix landed on `main` after the tag, and the chart reaches GHCR from exactly one place: three steps in `release.yml` each gated `if: github.event_name == 'push'`. A dispatch therefore packages, lints, verifies against the registry and deliberately stops, so a fix that lands after a tag has no route to the registry except a new tag — which is this one. The chart publishes as **`6.0.1`**: the 5.x → 6.x boundary is what announces the v13.0.0 behavioural break to anyone upgrading off `5.9.2`, and that boundary is crossed whichever 6.x publishes first, while the step up from the never-published `6.0.0` is a patch because the only thing that moved inside the chart is `appVersion` — no template, no default, no `values.yaml` key added, removed or renamed. `6.0.0` is not reused: it is already declared on `main` against a different `appVersion`, and unpublished is not the same as free. Everything below describes **v13.0.0**, which this patch does not change. **It was a major because it breaks behaviour, not a schema:** 75 state-changing API routes now require a permission they did not require before, and unlike the two patches earlier the same day, some of them refuse roles other than `viewer` — several surfaces move to `settings:write`, which `soc_analyst`, `soc_lead` and `threat_hunter` do not hold, so operators who are not tenant administrators will meet HTTP 403 where they previously got 200. The `### BREAKING` section of [`CHANGELOG.md`](CHANGELOG.md) lists every affected route, the permission it now takes, the roles that lose it, and what to do; `docs/openapi.yaml` is unchanged apart from one corrected description, which is exactly why the break is announced rather than left to be inferred from the spec. What it fixes is the debt the previous two patches measured but did not work down: `scripts/check_route_authz.py` reported **103 of 246 state-changing routes in `services/api` authenticating the caller and then checking nothing** — 42% of the write surface — and that figure is now **28**, every remaining route carrying a recorded reason. The worst module was MSSP, where `_require_own_child` established whose child a tenant was and nothing established whether the caller could act on it, so a `viewer` in a managing tenant could push a rule pack into a customer, grant itself a role inside one, or delete a critical detection from one. `community.py` had the moderation step gated and the thing being moderated open. Every permission used already existed and was already withheld from `viewer`, so there is no migration and no new permission to define. Upgrading is `make up`. v12.3.2 earlier the same day was the second security patch of the day: eleven more write routes were found declaring a permission that never ran. `Annotated[Any, require_permission("users:write")]` carries no `Depends()`, and FastAPI honours only `Annotated` metadata that is a `Depends` or a `FieldInfo` — everything else is silently dropped, so the check never happened and `current_user` degraded into a required query parameter. The tell was the status code: a `viewer` hitting these routes got **422, not 403**, which reads as a malformed request rather than a refusal. v12.3.1 was a security patch on top of v12.3.0, and the reason it exists separately is that a fix living on `main` is not a fix anyone has. Two reported advisories: a **critical** SQL injection in the osquery allowlist, where a denylist scanning rendered SQL for `--`, `/*` and `;` had never listed `'`, so a value could close a quoted literal and replace the WHERE clause — and reproducing it showed all six query templates were injectable, not the one reported, because the numeric parameters were never coerced either. Each parameter now declares what it is and anything else is refused rather than escaped, because an escaped path still reaches the endpoint, matches nothing, and reads to an operator as a clean host. The second is a **high** missing-authorization hole where every remediation write route authenticated and none authorized, so a `viewer` could raise the tenant to autonomy L4 and pre-approve a high blast-radius verb; the permission it needed already existed and was simply never enforced, which is the third time on this repository. The gate that should have caught it passed throughout, because it asks whether a route authenticates — so a second gate now asks whether it authorizes, and measures 103 of 246 state-changing routes making no authorization decision, as a ceiling that can only fall. Upgrading is `make up`. v12.3.0 (same day) closed six lettered sub-phases; its notes are below. None of the eight first-party npm/PyPI packages has been uploaded — the registries return 404 for all eight — and that remains credential-blocked rather than done. The latest GitHub release with notes and downloads: <https://github.com/beenuar/AiSOC/releases/latest>.
---
## What's new
-`VERSION` is `13.0.0`. The **v13.0.0** release (2026-09-29) works the
+`VERSION` is `13.0.1`. The **v13.0.1** patch (2026-09-29) ships no application
+change at all: it is the tag push that puts the corrected Helm chart into GHCR
+as **`6.0.1`**, the first 6.x a user can actually pull. v13.0.0 left the
+chart's own `version` at `5.9.2` while moving `appVersion`, and `helm push`
+replaces without a word, so that coordinate meant `v12.3.2` or `v13.0.0`
+depending on when it was fetched. The correction landed on `main` after the
+tag and could not reach the registry from there — `release.yml` gates its
+GHCR login, push and resolve steps on `github.event_name == 'push'`, so a
+dispatch packages and lints and then stops on purpose. `scripts/check_chart_version.py`
+now refuses the whole class before the push rather than after it. Everything
+in the rest of this section describes **v13.0.0**, which this patch leaves
+untouched.
+
+The **v13.0.0** release (2026-09-29) works the
identity-only route debt down from **103 to 28** across five reviewed changes
by resource area, using only permissions that already existed in
`ROLE_PERMISSIONS`. It is a major because the break is behavioural: several
diff --git a/VERSION b/VERSION
index 02161ca86..5cb7d8566 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-13.0.0
+13.0.1
diff --git a/apps/docs/static/data/scoreboard.json b/apps/docs/static/data/scoreboard.json
index 458087c90..8d193eb7a 100644
--- a/apps/docs/static/data/scoreboard.json
+++ b/apps/docs/static/data/scoreboard.json
@@ -4,8 +4,8 @@
"rows": [
{
"date": "2026-09-29",
- "agent_version": "v13.0.0",
- "commit_sha": "0f13e0fa",
+ "agent_version": "v13.0.1",
+ "commit_sha": "821bf0d1",
"substrate": true,
"eval_mode": "substrate-only",
"mitre_accuracy": 0.97,
diff --git a/apps/docs/static/llms-full.txt b/apps/docs/static/llms-full.txt
index 7b7fa49b9..cebe41667 100644
--- a/apps/docs/static/llms-full.txt
+++ b/apps/docs/static/llms-full.txt
@@ -3,7 +3,7 @@
<!-- Generated by scripts/generate_llms_txt.py. Do not edit by hand; run the script and commit the result. -->
> AiSOC is an open-source, self-hostable AI Security Operations Center, MIT
-> licensed and maintained by the AiSOC community. Version 13.0.0.
+> licensed and maintained by the AiSOC community. Version 13.0.1.
> Repository: https://github.com/beenuar/AiSOC. Documentation: https://beenuar.github.io/AiSOC/.
This file exists so an automated reader can answer questions about AiSOC
diff --git a/apps/docs/static/llms.txt b/apps/docs/static/llms.txt
index 32ccdb77f..e4deeebf4 100644
--- a/apps/docs/static/llms.txt
+++ b/apps/docs/static/llms.txt
@@ -6,7 +6,7 @@
> security telemetry from 84 first-party connectors, normalises it to OCSF,
> correlates it into alerts, and runs LLM agents that triage and investigate them.
> Every agent decision is written to an investigation ledger, so the reasoning
-> behind a verdict can be replayed step by step. MIT licensed. Current version 13.0.0.
+> behind a verdict can be replayed step by step. MIT licensed. Current version 13.0.1.
AiSOC is built and maintained by the AiSOC community. It is a complete SOC
platform rather than a single tool: ingestion, detection, correlation, agent
diff --git a/apps/web/package.json b/apps/web/package.json
index 74abf79a4..a6052400a 100644
--- a/apps/web/package.json
+++ b/apps/web/package.json
@@ -1,6 +1,6 @@
{
"name": "@aisoc/web",
- "version": "13.0.0",
+ "version": "13.0.1",
"private": true,
"description": "AiSOC SOC Console - Next.js 14 Frontend",
"license": "MIT",
diff --git a/infra/helm/aisoc/Chart.yaml b/infra/helm/aisoc/Chart.yaml
index 2e1c92006..18e389724 100644
--- a/infra/helm/aisoc/Chart.yaml
+++ b/infra/helm/aisoc/Chart.yaml
@@ -6,17 +6,26 @@ type: application
# move whenever anything in this directory does. v13.0.0 bumped `appVersion`
# and left this at 5.9.2, so `helm push` republished an existing chart version
# pointing at a different application — one immutable-by-convention coordinate
-# meaning two things depending on when it was pulled. A major here rather than
-# a minor because the application this chart installs has a breaking change
-# (see the v13.0.0 BREAKING section): a chart consumer running `helm upgrade`
-# gets it, whatever the templates did or did not do.
-version: 6.0.0
+# meaning two things depending on when it was pulled. The 5.x → 6.x boundary
+# is where that breaking application change (see the v13.0.0 BREAKING section)
+# is announced: a consumer moving off the published 5.9.2 crosses a major and
+# knows to read the notes, whatever the templates did or did not do.
+#
+# 6.0.0 itself never reached the registry — it landed on `main` after the
+# v13.0.0 tag, and only a tag push publishes — so the first pullable 6.x is
+# this one. A patch, because the only thing that moved inside this directory
+# is `appVersion` naming a patch release: no template, no default and no
+# values key added, removed or renamed, so no consumer's values file changes.
+# 6.0.0 is not reused for it. It is already declared on `main` against
+# appVersion v13.0.0, and a version that names two applications is the defect
+# the 5.9.2 overwrite taught; unpublished is not the same as free.
+version: 6.0.1
# Used as the image tag wherever a `tag:` in values.yaml is left empty, so
# it has to be a tag the registry actually holds — it read "5.2.0", which
# exists for no image, so every pod of a default install would have gone to
# ImagePullBackOff. scripts/check_published_images.py resolves it the same
# way the templates do and fails when it names nothing.
-appVersion: "v13.0.0"
+appVersion: "v13.0.1"
keywords:
- security