fullshield 1.4.0 → 1.6.0

package/CHANGELOG.md

@@ -2,6 +2,40 @@
 
 All notable changes to FullShield are documented here.
 
+## 1.6.0 — 2026-10-25
+
+### New/changed stable findings — may change CI results; use --update-baseline
+
+- Phase 507: `SAST-SQLI-001` also fires for `node:sqlite` `DatabaseSync.exec` / `prepare` and `bun:sqlite` `exec` / `prepare` when request data is concatenated or interpolated. A literal with placeholders and a separate bound argument does not fire. `child_process.exec` stays a command finding.
+- Phase 508: `SAST-CMDI-001` also fires for `Bun.spawn`, `Bun.spawnSync`, a Bun `$` template, and `Deno.Command` when the command or args are request data. A literal argv does not fire.
+- Phase 509: `SAST-EVAL-001` also fires for `evaluate` on `mathjs`, `expr-eval`, or `vm2`, and for `vm2` `VM` / `NodeVM` when the code argument is not a string literal. An unrelated `evaluate()` does not fire. `mathjs.eval('1 + 2')` stays silent.
+
+### Experimental
+
+- Phase 510: `DEPS-RSC-001` behind `FULLSHIELD_EXPERIMENTAL_RSC` (`scan --experimental-rsc`). Exact React Server Components versions 19.0.0, 19.1.0, 19.1.1, and 19.2.0 in a lockfile. A canary does not fire. Off by default.
+- Phase 511: `CONFIG-COOLDOWN-001` behind `FULLSHIELD_EXPERIMENTAL_COOLDOWN` (`scan --experimental-cooldown`). A release-age gate shorter than three days, in that tool's unit. No cooldown file means no finding. Off by default.
+- Phase 512: `SUPPLY-IMPLANT-001` behind `FULLSHIELD_EXPERIMENTAL_IMPORTIMPLANT` (`scan --experimental-import-implant`). `setup_bun.js` or `bun_environment.js`, including install scripts and `node_modules` up to 2000 manifests. Off by default.
+- Phase 513: `DEPS-VERINFLATE-001` behind `FULLSHIELD_EXPERIMENTAL_VERINFLATE` (`scan --experimental-ver-inflate`). Lockfile major 9999+ or patch 999 or 9999. Off by default.
+- Phase 514: `AGENT-UNICODE-001`, `AGENT-INJECT-001`, `AGENT-EXFIL-001`, and `AGENT-SHELL-001` behind `FULLSHIELD_EXPERIMENTAL_AGENTRULES` (`scan --experimental-agent-rules`). Named agent instruction files only. Off by default.
+- Phase 515: `SAST-AGENTSPAWN-001` behind `FULLSHIELD_EXPERIMENTAL_AGENTSPAWN` (`scan --experimental-agent-spawn`). A coding agent launched with a permission bypass in the same call. Off by default.
+- Phase 516: `MCP-006`, `MCP-007`, and `MCP-008` behind `FULLSHIELD_EXPERIMENTAL_MCPPOISON` (`scan --experimental-mcp-poison`). Secret elicitation, a shell or instruction default, and a duplicated tool name. `MCP-001`–`MCP-005` are unchanged. Off by default.
+- Phase 517: `CONFIG-VITEFS-001` behind `FULLSHIELD_EXPERIMENTAL_VITEFS` (`scan --experimental-vite-fs`). `server.fs.strict: false` or a filesystem root in `server.fs.allow`. Off by default.
+- Phase 518: `SAST-HOST-001` behind `FULLSHIELD_EXPERIMENTAL_HOSTPOISON` (`scan --experimental-host-poison`). Host or X-Forwarded-Host copied into an absolute redirect or mail URL. Off by default.
+- Phase 519: `SAST-MARKUP-001` behind `FULLSHIELD_EXPERIMENTAL_MARKUP` (`scan --experimental-markup`). markdown-it `html: true` or marked, rendering request data. Off by default.
+- Phase 520: `SAST-TRUSTHOST-001` behind `FULLSHIELD_EXPERIMENTAL_TRUSTHOST` (`scan --experimental-trust-host`). Auth.js `trustHost: true`. Off by default.
+
+## 1.5.0 — 2026-10-24
+
+### New/changed stable findings — may change CI results; use --update-baseline
+
+- Phase 503: one object field and one nested function are part of the default scan. The nested function carries request data, not `process.argv`. A field counts on a read after the assignment. `FULLSHIELD_EXPERIMENTAL_FLOW2` and `scan --experimental-flow2` stay accepted and do not turn this off. Against 1.4.0, NodeGoat gains `SAST-XSS-001` at `app/views/allocations.html:15` and `app/views/login.html:110`. vulnerable-app stays 38, clean-app stays 0, and juice-shop, dvna, goof, and vulnerable-node stay the same.
+- Phase 504: ORM raw queries, `` exec(`…${x}`) ``, `spawn` of `sh -c`, execa `shell: true`, and zx `$.raw` are part of the default scan. `FULLSHIELD_EXPERIMENTAL_SAST_GAPS` and `scan --experimental-sast-gaps` stay accepted and do not turn this off. No new rule id.
+
+### Experimental
+
+- Phase 505: `SAST-LDAP-001` behind `FULLSHIELD_EXPERIMENTAL_LDAP` (`scan --experimental-ldap`). A string-literal filter does not fire. Off by default.
+- Phase 506: `CONFIG-CORS-002` behind `FULLSHIELD_EXPERIMENTAL_CORSREFLECT` (`scan --experimental-cors-reflect`). A fixed origin does not fire. `origin: '*'` with credentials stays `CONFIG-CORS-001`. Off by default.
+
 ## 1.4.0 — 2026-10-24
 
 ### Experimental

package/README.md

@@ -1,6 +1,6 @@
 # FullShield
 
-**v1.4.0.** One install for Node.js/TypeScript backends: detect → classify → report → notify → fix → gate (CI `--fail-on` + runtime monitor). It does not stop attacks by itself.
+**v1.6.0.** One install for Node.js/TypeScript backends: detect → classify → report → notify → fix → gate (CI `--fail-on` + runtime monitor). It does not stop attacks by itself.
 
 **Headline: dependency reachability.** A lockfile CVE is not the same as an unused package. FullShield classifies each OSV hit as `reachable` (imported and called), `imported-not-called` (−1), `reachable-via-dependency` (lockfile path from an imported direct dep — **severity kept**, path shown, e.g. `express > body-parser > qs`), `not-imported` (−2, only when no such path exists), or `reachability-unknown` (severity kept) when the lockfile format has no parsed dependency edges. DevDependency-only vulns are `info` with a `dev-only` note unless production sources import the package. Pair it with `fullshield sbom --format cyclonedx`.
 
@@ -10,7 +10,7 @@ npx fullshield scan --json --fail-on high
 npx fullshield scan --sarif fullshield.sarif
 ```
 
-1.4.0 is the version of this tree. The stable contract is [docs/stability.md](docs/stability.md). The 1.0.1 release moved values that only experimental flags produce out of Stable. Experimental flags stay off unless you set them. Do not treat a scan as a penetration test. Reviewed recall for untouched apps is not published. The stored Next.js cold scan is 71.7 seconds, above the 60-second target.
+1.6.0 is the version of this tree. The stable contract is [docs/stability.md](docs/stability.md). The 1.0.1 release moved values that only experimental flags produce out of Stable. Experimental flags stay off unless you set them. Do not treat a scan as a penetration test. Reviewed recall for untouched apps is not published. The stored Next.js cold scan is 71.7 seconds, above the 60-second target.
 
 **Supported runtime is Node.js >= 22.13.** `package.json` `engines` and `doctor` enforce that floor. `node:sqlite` runs without `--experimental-sqlite` from 22.13.0 (the module is still experimental and is loaded only by history/confirm, not by `--import` register).
 
@@ -85,7 +85,7 @@ Experimental RASP structure-change scoring on the vendored PayloadsAllTheThings
 
 Path 51/51 and SSRF 39/39 on the same corpus. Benign corpus stays 0 blocked. Prefer `mode: 'monitor'`.
 
-These fixtures are self-authored. Numeric SQL (`WHERE id = ` + input), ORM raw-query APIs (Prisma/Sequelize/Knex/TypeORM/Drizzle/`$where`), and template-literal shell (`exec(\`…${x}\`)`, `spawn('sh', ['-c', …])`, execa `shell: true`, zx `$.raw`) are covered in the attack corpus. Extra SAST sinks for those ORM / `sh -c` / execa / zx shapes stay behind `FULLSHIELD_EXPERIMENTAL_SAST_GAPS=1` (or `scan --experimental-sast-gaps`) so default `scan` behavior does not change. Mongo `$where` strings are **stable** SAST (Juice Shop track-order / reviews).
+These fixtures are self-authored. Numeric SQL (`WHERE id = ` + input), ORM raw-query APIs (Prisma/Sequelize/Knex/TypeORM/Drizzle/`$where`), and template-literal shell (`exec(\`…${x}\`)`, `spawn('sh', ['-c', …])`, execa `shell: true`, zx `$.raw`) are covered in the attack corpus. Those ORM and shell sinks are part of the default scan. `FULLSHIELD_EXPERIMENTAL_SAST_GAPS` and `scan --experimental-sast-gaps` stay accepted and do not turn them off. Mongo `$where` strings are **stable** SAST (Juice Shop track-order / reviews).
 
 In-scope public-bench patterns we still do not claim (one line each):
 
@@ -226,9 +226,22 @@ app.use(shieldGuard({ mode: 'monitor', authRoutes: ['/login'] }));
 | `agent` | **Experimental** remediation agent (`FULLSHIELD_EXPERIMENTAL_AGENT=1`). Mock/localhost GitHub API + local git remote only — never a real network. Never merges, never pushes to `main`/`master`. `--max-prs` (default 3). |
 | `scan --changed` / `--since` | **Experimental** incremental scan (`FULLSHIELD_EXPERIMENTAL_PLATFORM=1`). |
 | `scan --diff` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_DIFFSCAN=1` or `--experimental-diffscan`). Scans git-changed JS/TS files, the files they import by a relative literal, and the files that import them the same way. One hop. A sink in an unchanged importer still reports when the changed file returns the request value. A git failure is `FS-E018` and does not scan the rest of the tree. Off, or `scan` without `--diff`, matches today's scan. |
-| `scan --experimental-flow2` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_FLOW2=1`). Existing injection rules also fire when request data reaches the sink through one object field (`obj.a`) or one nested function that closes over the tainted name. `obj.b` stays clean. `JSON.parse` does not clear taint. Handlebars `{{ }}`, `res.json`, and a `url.includes` allowlist helper stay unflagged. |
+| `scan --experimental-flow2` | **Stable.** One object field (`obj.a`) and one nested function are part of the default scan. `FULLSHIELD_EXPERIMENTAL_FLOW2` and this flag stay accepted and do not turn that off. `obj.b` stays clean. `JSON.parse` does not clear taint. Handlebars `{{ }}`, `res.json`, and a `url.includes` allowlist helper stay unflagged. |
 | `dev` | **Experimental** save loop (`FULLSHIELD_EXPERIMENTAL_DEV=1`). Watches the project and prints only new and resolved findings. |
-| `scan --experimental-sast-gaps` | **Experimental** extra SAST sinks: ORM raw SQL, mongoose `$where`, `spawn('sh', ['-c'])`, execa `shell: true`, zx `$.raw` (`FULLSHIELD_EXPERIMENTAL_SAST_GAPS=1`). |
+| `scan --experimental-sast-gaps` | **Stable.** ORM raw SQL, `spawn('sh', ['-c'])`, execa `shell: true`, and zx `$.raw` are part of the default scan. `FULLSHIELD_EXPERIMENTAL_SAST_GAPS` and this flag stay accepted and do not turn that off. |
+| `scan --experimental-ldap` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_LDAP=1`). `SAST-LDAP-001`: `search`, `compare`, `modify`, or `bind` on an LDAP client when the filter or DN is request data. A string-literal filter does not fire. Off by default. |
+| `scan --experimental-cors-reflect` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_CORSREFLECT=1`). `CONFIG-CORS-002`: `Access-Control-Allow-Origin` is the request `Origin` and credentials are allowed on that response. A fixed origin does not fire. `origin: '*'` with credentials stays `CONFIG-CORS-001`. Off by default. |
+| `scan --experimental-rsc` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_RSC=1`). `DEPS-RSC-001`: a lockfile resolves `react-server-dom-webpack`, `react-server-dom-parcel`, or `react-server-dom-turbopack` to 19.0.0, 19.1.0, 19.1.1, or 19.2.0. A canary or any other version does not fire. No lockfile means no finding. Off by default. |
+| `scan --experimental-cooldown` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_COOLDOWN=1`). `CONFIG-COOLDOWN-001`: `.npmrc`, `pnpm-workspace.yaml`, `.yarnrc.yml`, or `bunfig.toml` is present and the release-age gate is missing or shorter than three days. npm is days, pnpm is minutes, yarn is a duration, bun is seconds. A project with none of those four files does not fire. Off by default. |
+| `scan --experimental-import-implant` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_IMPORTIMPLANT=1`). `SUPPLY-IMPLANT-001`: `setup_bun.js` or `bun_environment.js` is in the scan, named by a preinstall/install/postinstall script, or present under `node_modules` (capped at 2000 manifests, no symlinks). Off by default. |
+| `scan --experimental-ver-inflate` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_VERINFLATE=1`). `DEPS-VERINFLATE-001`: a lockfile version whose major is 9999 or greater, or whose patch is 999 or 9999. `1.2.3` does not fire. Off by default. |
+| `scan --experimental-agent-rules` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_AGENTRULES=1`). `AGENT-UNICODE-001`, `AGENT-INJECT-001`, `AGENT-EXFIL-001`, and `AGENT-SHELL-001` on `AGENTS.md`, `CLAUDE.md`, `.github/copilot-instructions.md`, `.cursor/rules/**`, `.windsurf/rules/**`, and `SKILL.md` under `.cursor/`, `.claude/`, or `.agents/`. A short guideline with none of those markers does not fire. Off by default. |
+| `scan --experimental-agent-spawn` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_AGENTSPAWN=1`). `SAST-AGENTSPAWN-001`: `exec`, `execSync`, `spawn`, `spawnSync`, `Bun.spawn`, `Bun.spawnSync`, or `new Deno.Command` launches `claude`, `codex`, or `gemini` with `--dangerously-skip-permissions`, `--permission-mode bypassPermissions`, or `--yolo` in the same call. The program without the flag does not fire. Off by default. |
+| `scan --experimental-mcp-poison` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_MCPPOISON=1`). JSON that contains `mcpServers`. `MCP-006` asks for a password, token, secret, or API key. `MCP-007` puts `ignore previous` or a shell marker in a schema default. `MCP-008` repeats one tool name on two servers in one file. One server with a literal description does not fire. `MCP-001`–`MCP-005` stay on `FULLSHIELD_EXPERIMENTAL_MCPCFG`. Off by default. |
+| `scan --experimental-vite-fs` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_VITEFS=1`). `CONFIG-VITEFS-001`: a Vite config sets `server.fs.strict` to false or `server.fs.allow` to `/` or a drive root. Omitted `server.fs` does not fire. Off by default. |
+| `scan --experimental-host-poison` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_HOSTPOISON=1`). `SAST-HOST-001`: `Host` or `X-Forwarded-Host` is copied into an absolute URL passed to `redirect`, `location`, `sendMail`, or a `Location` header. A fixed host does not fire. Off by default. |
+| `scan --experimental-markup` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_MARKUP=1`). `SAST-MARKUP-001`: `markdown-it` is constructed with `html: true` and `render` is called with request data, or `marked` / `marked.parse` is called with request data. The file must import that package. Literal markdown does not fire. Off by default. |
+| `scan --experimental-trust-host` | **Experimental** (`FULLSHIELD_EXPERIMENTAL_TRUSTHOST=1`). `SAST-TRUSTHOST-001`: a file that mentions `next-auth`, `@auth/`, or `NextAuth` sets `trustHost: true`. `false` or omitted does not fire. Off by default. |
 | `scan --experimental-frameworks` | **Experimental** Next.js / NestJS / Hono / Koa taint sources and AUTH (`FULLSHIELD_EXPERIMENTAL_FRAMEWORKS=1`). Default `scan` is unchanged. Runtime adapters: `fullshield/runtime/koa`, `fullshield/runtime/hono` (optional peers). |
 | `scan --experimental-classes` | **Experimental** extra vuln classes: ReDoS, deser, XXE, upload, cookies, CSRF, mass assignment, timing compare, hash cost, GraphQL, trust-proxy (`FULLSHIELD_EXPERIMENTAL_CLASSES=1`). Default `scan` is unchanged. |
 | `scan` crypto v2 | **Experimental** (`FULLSHIELD_EXPERIMENTAL_CRYPTO2=1`). `SAST-CIPHER-001`, `SAST-CIPHER-002`, `SAST-RSA-001`, and `SAST-JWT-002`: ECB and deprecated ciphers, literal IVs, literal symmetric keys, short RSA or secp192 keys, and JWTs with no expiry or a short HMAC secret. Off by default. |