Back to skill

Security audit

FTA Studio 重建

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report potentially destructive installation and process stopping, along with issues that could affect safety-analysis results; some checks were incomplete.

Consider avoiding installation until the destructive replacement and broad process termination risks are addressed, and the reported analysis-result issues are independently reviewed. The scan is incomplete: it reports 49% component coverage, with partial analysis and degraded static checks, so unreported behavior has not been ruled out.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
.
Finding

Recursive installation deletion lacks an enforced confirmation gate

Content
View full analysis

Impact

The entire existing installation directory is deleted, including any locally modified code or additional files stored there. Installation removes it before validating and successfully copying a complete replacement. The target is limited to the current user's LOCALAPPDATA/FTA Studio directory; arbitrary filesystem deletion is not established.

T09 · Insecure Skill Coding Practices

Warning
Location
.
Finding

Maintenance forcibly terminates unrelated Node launchers

Content
View full analysis

Impact

Unrelated Node applications can be forcibly stopped, interrupting their work. Impact is limited by the invoking user's process-termination permissions; privilege escalation and cross-user termination are not established.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (29)

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 58)May include surrounding context.

md
5. **改完必须同步到用户实际运行的那一份**:用户跑的是 `%LOCALAPPDATA%\FTA Studio\app`,不是你改的源码目录。跑 `scripts/sync-installed.ps1 -SourceDir <项目> -Restart`,否则一切努力都白费(见 `pitfalls.md` §9)。

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 107)May include surrounding context.

md
该脚本用 `Input.dispatchMouseEvent` 产生**真实鼠标序列**(含 `clickCount=2` 的双击)。见 `references/pitfalls.md` 第 1 条:用 `dispatchEvent(new MouseEvent('dblclick'))` 会**假通过**。

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 126)May include surrounding context.

md
| `references/pitfalls.md` | **遇到诡异现象先看这里**:已踩过的坑及根因 |

Memory Manipulation

High
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.

Content

Scanner excerpt · assets/source/public/app.js (reported line 771)May include surrounding context.

js
for (const k of (kids.get(nid) || [])) { ids.push(k); walk(k); }
  };
  walk(id);
  for (const x of ids) delete state.offsets[x];
  render();
  renderProps();
  focusNode(id);

Memory Manipulation

High
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.

Content

Scanner excerpt · assets/source/public/app.js (reported line 1045)May include surrounding context.

js
function deleteNode(id) {
  state.tree.nodes = state.tree.nodes.filter((n) => n.id !== id);
  state.tree.connections = state.tree.connections.filter((c) => c.from !== id && c.to !== id);
  delete state.offsets[id];
  if (state.flash === id) state.flash = null;
  if (state.selected === id) state.selected = null;
  render(); renderProps(); renderOutline();

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · assets/source/src/launch.mjs (reported line 21)May include surrounding context.

js
// ---- 0) 让本机回环流量绕过任何 HTTP 代理 ----
// Node 24 的全局 fetch 会读取 HTTP_PROXY/HTTPS_PROXY;企业代理一旦拦截 127.0.0.1,
//  readiness 探测与浏览器都会"无法访问此页面"。本工具无任何外部请求,故直接清掉代理变量。
for (const k of Object.keys(process.env)) {
  if (/proxy/i.test(k)) delete process.env[k];
}
process.env.no_proxy = '127.0.0.1,localhost,::1';

Page-view generation can rewrite the original tree's connections despite documenting that pagination leaves the original tree intact (L231–L233).

High
Category
Excessive Agency
Confidence
99% confidence
Finding

Filtering tree.connections creates a new array but retains references to the original connection objects; assigning conn.to therefore modifies the original fault tree. Generating a page view can redirect analytical connections to display-only transfer nodes, compromising subsequent fault-tree calculations or saved models—a significant integrity risk in safety-analysis software.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, network.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Installation recursively deletes an existing installation without disclosing that destructive replacement.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The installer recursively deletes the existing application directory before validating the replacement package, without confirmation or backup. Reinstallation can therefore destroy locally stored files or leave the application unavailable if copying fails; the fixed application-specific target limits the deletion scope.

Content

No source excerpt is available for this finding.

The uninstaller can forcibly terminate unrelated Node.js applications (L10–L12).

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

Matching every node.exe process whose command line contains launch.mjs does not establish that the process belongs to FTA Studio. Uninstalling can forcibly terminate unrelated applications accessible to the invoking user, causing service interruption or loss of unsaved work.

Content

No source excerpt is available for this finding.

The FTA Studio stop script can forcibly terminate unrelated Node.js applications (L09).

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The stop command identifies targets using only node.exe and a generic launch.mjs command-line match. Consequently, an ordinary application-stop operation can forcibly terminate unrelated Node.js applications, potentially interrupting services or discarding unsaved state.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · assets/source/AGENT_API.md (reported line 74)May include surrounding context.

md
node src/cli.mjs serve --port 8777

# 分析
curl -X POST http://127.0.0.1:8777/api/analyze \
  -H "Content-Type: application/json" \
  -d '{"dsl":"TOP 顶事件\n  OR\n    BASIC A p=0.01"}'

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · assets/source/AGENT_API.md (reported line 298)May include surrounding context.

bash
# 计算分页方案
curl -X POST http://127.0.0.1:8777/api/paginate \
  -H "Content-Type: application/json" \
  -d '{"dsl":"...","options":{"maxNodes":24}}'

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · assets/source/AGENT_API.md (reported line 353)May include surrounding context.

bash
# 导出
curl -X POST http://127.0.0.1:8777/api/export \
  -H "Content-Type: application/json" \
  -d '{"dsl":"...","options":{"format":"faulttreeplus"}}'

The documented exact importance analysis returns incorrect RAW and RRW metrics.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The exported function labels p1 as RAW and (1 − p0)/p0 as RRW, whereas the documented ratios require baseline top-event probability Q: RAW = p1/Q and RRW = Q/p0. Consumers of this helper can receive misleading safety-analysis metrics, although the supplied source does not establish whether production reports call it.

Content

No source excerpt is available for this finding.

The documented subset-absorption guarantee is contradicted by pruning only within matching lowest-bit buckets (L331–L383).

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

A subset need not share its superset's lowest set bit: for example, binary 10 is a subset of 11, but these masks enter different buckets and are never compared. This leaves nonminimal cut sets in results and can distort cut-set-derived statistics or importance measures, which matters particularly in a safety-analysis tool.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

House-event probabilities invert the documented TRUE/FALSE meaning.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The assignment maps an explicitly FALSE house event or prob=0 to probability 1, while otherwise assigning 0, reversing the documented Boolean meaning. This corrupts the probability map and any calculations that consume it; in a fault-tree safety tool, reversed scenario conditions can produce materially misleading risk assessments.

Content

No source excerpt is available for this finding.

Documented MCUB and Esary-Proschan calculations are only an uncorrected rare-event sum.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The code reports the rare-event sum as MCUB and as the top-event upper bound without implementing the advertised method. An untruncated sum is a conservative union bound, not inherently an underestimate, but when cut sets are truncated this value is not necessarily an upper bound on the complete model, making the reported assurance potentially unsafe.

Content

No source excerpt is available for this finding.

Intermediate probabilities never use the documented BDD-exact calculation.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

Intermediate probabilities use local gate formulas rather than BDD evaluation, implicitly treating child events as independent even when branches share basic events. These values are exposed as node probabilities and heatmap inputs, so the interface can display incorrect safety-relevant results despite an accurate BDD top-event result.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Mandatory installed-copy synchronization and restart lack a warning about overwrites and interruption.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The guidance makes synchronization into the installed application and restart mandatory after every change, extending ordinary source editing into modification of a separate running installation. Without checking consent, local modifications, or unsaved work, following this instruction can overwrite installed changes or interrupt an active session; the script implementation is not provided here to establish additional safeguards.

Content

No source excerpt is available for this finding.

Uninstall guidance can terminate unrelated Node applications.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

Matching Node processes solely by launch.mjs does not identify FTA Studio uniquely, since unrelated applications can use the same launcher filename. Following this uninstall guidance can terminate other applications accessible to the user, causing service interruption or loss of unsaved work.

Content

No source excerpt is available for this finding.

Restart terminates all matching launch.mjs Node processes, not just the installed FTA Studio service.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

With -Restart, the script forcibly terminates every accessible node.exe process whose command line contains launch.mjs, without verifying that it belongs to the selected FTA Studio installation. Unrelated applications or other installations can therefore be stopped, causing service disruption or loss of in-progress work; running the script elevated increases the potential scope.

Content

No source excerpt is available for this finding.