Back to skill

Security audit

hd2-mod-authoring

Security checks for vulnerabilities and agentic risk

Overview

This is a high-risk but openly documented Helldivers 2 mod-authoring skill that stays aligned with its stated purpose.

Install only if you want an agent to help edit HD2 mod files, deployment metadata, and diagnostic addons. Before allowing writes, confirm the target path, keep backups, close the mod manager as instructed, and protect or delete memory-dump outputs after use.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (12)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared description presents a large multifunction HD2 mod-installation, troubleshooting, editing, verification, and malware-vetting tool. The supplied code instead only packs a single Lua script into a specific patch_0 container, creates empty sidecar files, and builds a zip archive with fixed files. Its assertions merely validate the generated archive structure, not gameplay data, mod conflicts, malware, or runtime logs. This is a materially different primary purpose, so the description does not accurately represent the code chunk.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The declared purpose describes a large diagnostic and mod-analysis toolchain for many Helldivers 2 modding scenarios, including security vetting, conflict analysis, editing, packing, and verification against game logs. The supplied code only implements the 'pack' portion for one specific addon: it constructs a binary patch container for refscan.lua, emits empty sidecar files, and zips fixed files into a distributable archive. It does not inspect other tools for malware, parse logs, analyze conflicts, validate magazine/explosion data, translate HUD text, or perform installation/repair tasks. While packing is mentioned in the description, the actual code's primary behavior is much narrower than the declared functionality, so the description does not accurately represent what this code chunk actually does.

Content

No source excerpt is available for this finding.

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · SKILL.md (reported line 361)May include surrounding context.

md
it outright). A mod that *fails safe* — its keyed lookup requires "exactly one
  layout fits" plus a plausibility ceiling, so drift no-ops cleanly instead of
  corrupting — is still broken, just silently (Realistic Weapons is this class).
  Deriving an entity hash uses the fingerprint rule above, not the projectile
  type number by itself. diag-5 already writes the join to `refscan_pwbuckets.txt`
  (`references/memory-census.md`). A census without that file is the older dump
  and cannot be joined offline. Do not guess a fingerprint, and do not extend

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding

The skill describes extensive file read/write behavior against live mod directories, deployment snapshots, game data files, and package archives, but declares no explicit tool scope or permissions boundary. In an agent setting, that omission is dangerous because it leaves broad filesystem mutation implied by natural-language instructions rather than constrained by enforceable policy, increasing the chance of unintended or unsafe writes.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
83% confidence
Finding

The skill embeds extensive stateful operational guidance tied to prior runs, prior revisions, deployed copies, and persistent local files, encouraging the agent to rely on mutable session and filesystem state across tasks. In practice this can cause unsafe decisions based on stale deployment state or prior artifacts, especially when the skill also instructs edits to installed files and verification against logs.

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: hd2-mod-authoring
description: "Helldivers 2 mod, Arsenal install, Bingus lua patch, Wwise sound-bank swap, Nexus unlink, an HD2 Mod Manager install that dies on the manifest Name, translating on-screen text in a glyph-atlas HUD, drawing a live HUD overlay with the game font, a bundled .patch_0 whose type at 0x50 is not Lua, the whole HUD freezes after the stratagem list and death does not clear it, Bingus reports more than 200 local variables or more than 60 upvalues, the realistic log says 4 spare magazines while the gun still shows 6, the weapon search ends a round at 256 MB after ProjectileWeapon, the magazine write refuses because capacity reads 0, the log says explosion table not found, the log says explosion on impact 0 -> 388 while a close SAFE shot still detonates, black squares after the impact id moves onto a spare explosion slot, the Dominator thumbnail still lists only the 380 m/s specs, a pytest check of the magazine index or explosion tail without launching the game, or statically vetting a downloaded HD2 mod manager/tool for malware before running it: classify the install, conflict-check writers of the same value, then edit, pack, and verify from the game log."
homepage: "https://github.com/Cavadus/hd2-mod-authoring"
---

Session Persistence

Medium
Category
Rogue Agent
Confidence
91% confidence
Finding

This section explicitly directs modification of persistent config and deployment metadata such as hd2a_data.json, deployment_snapshot.json, content hashes, and deployed patch files. That is dangerous in an agent context because it normalizes direct tampering with application state across sessions, which can corrupt mod-manager state, hide unauthorized changes, or cause future tasks to trust manipulated metadata.

Content

Scanner excerpt · SKILL.md (reported line 183)May include surrounding context.

md
The operator still quits Arsenal, opens it again, and deploys before a game test. `deployed: false` is not a deployed mod.

Editing `~/.config/hd2arsenal/mods/<folder>/` does not change what the game loads. Deploy copies each enabled patch into `Helldivers 2/data/` as `9ba626afa44a3aa3.patch_<N>` (plus `.gpu_resources` / `.stream` siblings). `N` is assigned at deploy time and is not stable; `patch_0` is only the first slot, not the Lua file. `~/.config/hd2arsenal/deployment_snapshot.json` `deployedFiles` lists `{name, size, modified}` for the last deploy. Match the mod's patch by size, then by a unique string. On the 2026-10-07 23:45 deploy, HD2 HUD Plus 0.2.2 (721977 bytes) was `patch_95`; `data/9ba626afa44a3aa3.patch_0` was 15672 bytes and a different file. `hd2a_data.json` `enabled: true, deployed: true, changed: false` does not prove the data file matches the mods folder — compare bytes. After an edit, copy the same bytes onto that `patch_N`, or have the operator redeploy. Bingus reads the data file once, at process start. Before saying a restart is enough, confirm the new marker is in the data file, not only in the mods folder. Toggling the mod, or an option under it, while the game is open does not reload it. `contentHash` on both the library record and the profile record is the sha256 of that mod's patch-asset bytes sorted as raw buffers (`.patch_0`, `.gpu_resources`, and `.stream` for HUD Plus matched the stored hash). Recompute it after the edit and write it on both records. Set the matching `deployedFiles` entry's `modified` to the data file's mtime in milliseconds. Leave `changed: false` and `deployed: true` only when those data-file bytes already match. Do not Nexus-update over a local patch of that bundle; a re-download replaces the edit. 0.2.2 (721977 bytes, `patch_95`) was current on 2026-10-07. A later library record named HUD Plus 0.2.4, so match the live file before reapplying that edit.

An option can be checked while the mod record's own `enabled
...[truncated 25 chars]

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

This markdown file states that the addons read game memory and dump results to %LOCALAPPDATA%/Hd2ProjRecon/, but it does not warn users that the output may contain sensitive memory-derived information or advise caution in handling those dumps. For a skill description, memory inspection plus file output can affect privacy or local data handling and should be disclosed clearly as a user warning.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The header documentation states this is a "READ-ONLY value-scan" and explicitly says it "Writes NOTHING to game memory," which conveys a no-write intent. However, the script creates a directory and writes a dump file containing candidate records to %LOCALAPPDATA%/Hd2ProjRecon/fmscan_records.txt, so the documentation's practical no-write implication is contradicted by actual file-system side effects.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The file header explicitly states the script is 'READ-ONLY' and 'Writes nothing to game memory', but the implementation creates a directory under %LOCALAPPDATA% and writes multiple text and binary dump files there. While it does not appear to modify game memory, the misleading claim reduces operator awareness and can cause users to run a memory-dumping tool without understanding that it persists potentially sensitive process data to disk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

These lines describe replacing Chinese text/assets with English equivalents, including that the shipped nicknames and destroyed label are English and that a UTF-8 scan found no CJK. This is a natural-language locale policy issue because the document prescribes a specific language outcome rather than offering a user-selectable language or documenting a justified region-specific constraint.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
79% confidence
Finding

Lines L003-L005 describe the addon as a 'Read-only memory census' that 'Writes nothing to game memory,' which frames it as purely observational. Later content explicitly describes dumping raw binary via io.open and consulting per-enemy patch logic files under /tmp; while this does not contradict 'no game-memory writes,' it does contradict a plain-language read-only characterization if interpreted as no side effects at all.

Content

No source excerpt is available for this finding.

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · SKILL.md (reported line 637)May include surrounding context.

md
independent — set them separately, and you can swap which icon goes on which
   round freely.

   - **Label is a RAW localisation string id**, not limited to the
     `modePresentation` catalog's offered labels. Any string in the game's
     `.strings` database resolves. To make the wheel read a specific word, find
     that word's id: extract the strings and read its `Key`: