Back to skill

Security audit

4d-prototype

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence because scanners report privacy and process-disruption risks, while analysis of some files was incomplete.

Consider the reported spreadsheet data exposure and the possibility of stopping unrelated browser sessions before installing. The scan is partial: 24 of 34 components were fully inspected, and some files or security checks could not be fully evaluated, so the evidence is incomplete.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (16)

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 74)May include surrounding context.

md
s "is the process right?" into checks a machine can run. Keep invariant tests (`flow.test.ts`, see flow-orchestration.md §2) and scenario tests green on every c

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 95)May include surrounding context.

md
tes document, README | First deliverable version | [references/verification.md](references/verification.md) |

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 190)May include surrounding context.

md
- Checklist: [references/verification.md](references/verification.md).

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 248)May include surrounding context.

md
| [references/verification.md](references/verification.md) | Verifying every milestone; writing headless-browser scripts; debugging unreadable screenshots or st

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 96)May include surrounding context.

md
totype whose timeline plays with one click | [references/flow-orchestration.md](references/flow-orchestration.md), [assets/flow-kit/](assets/flow-kit/README.md)

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 151)May include surrounding context.

md
he skill — the fourth dimension. The spec is [references/flow-orchestration.md](references/flow-orchestration.md). A reusable TypeScript reference implementatio

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 247)May include surrounding context.

md
| [references/flow-orchestration.md](references/flow-orchestration.md) | Implementing or changing any part of the time axis (the 4D orchestration layer) |

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 189)May include surrounding context.

md
- Full-resolution screenshots, long autoplay runs, console error collection: drive headless Chrome with `scripts/cdp_run.mjs`.

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 252)May include surrounding context.

md
| `scripts/cdp_run.mjs` | Full-resolution 1920×1080 screenshots, polling long flows, collecting console errors |

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/flow-kit/src/state/session.ts (reported line 6)May include surrounding context.

ts
/**
 * The signed-in user (demo role). In production, single sign-on / a portal cookie provides the employee id and the backend looks up
 * organization, role and data scope in the user–organization table; the prototype simulates this with the demo role switcher in the top bar.
 * The example is purchase request approval; rewrite RoleKey / USERS per project.
 */
export type RoleKey = 'applicant' | 'approver' | 'finance' | 'admin';

Context Leakage

High
Category
Data Exfiltration
Confidence
85% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · references/pages-and-ux.md (reported line 80)May include surrounding context.

md
| Pattern | Purpose | Key points | Common in |
|---|---|---|---|
| Form and validation | Where input comes from | Field groups, required and format checks, a "fill example" button (the demo cursor clicks it) | Approvals, admin |
| Upload dialog | Where input comes from | Several built-in demo files (success / wrong category / non-standard format); a real file can also be picked and is parsed only locally; a "file format" link next to it | Data processing |
| List + detail | Finding objects | Filters, status pills, to-dos; the detail is a document-style field view with problem fields in red | All |
| State transitions / approval history | Where the object is now | Timeline: state, actor, time, comment; current node highlighted; reject or send-back branches visible | Approvals, tickets |
| Staged processing | What the system is doing | Parse → normalize → rules → statistics → summary, each stage labeled with actor and duration; can compare run modes (batch / per record) | Data processing

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

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.

The port-conflict workaround terminates matching processes without warning about disrupting other browser sessions.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The recommended pkill command terminates every accessible process whose command line matches the debugging-port string, without verifying that the current test run owns it. An unrelated browser or automation session using that port could be interrupted, potentially losing in-progress work; routine milestone testing makes accidental execution plausible.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/pdf_to_text.py (reported line 27)May include surrounding context.

python
pages = []
    i = 1
    while True:
        r = subprocess.run(['pdftotext', '-layout', '-f', str(i), '-l', str(i), src, '-'], capture_output=True, text=True)
        if r.returncode != 0:
            break
        pages.append(r.stdout)

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/pdf_to_text.py (reported line 70)May include surrounding context.

python
def via_pdfkit(src):
    if sys.platform != 'darwin' or not shutil.which('osascript'):
        return None
    r = subprocess.run(['osascript', '-l', 'JavaScript', '-e', JXA, src], capture_output=True, text=True)
    if r.returncode != 0:
        return None
    try:

Sensitive-column detection does not enforce the documented privacy guarantees.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The sensitivity heuristic misses identifiers under unrecognized headers, formatted phone or account numbers without eight consecutive digits, and sensitive values appearing beyond the first 2,000 entries. Low-cardinality columns that evade detection are printed verbatim and included in JSON output, so real customer data can enter agent context or saved artifacts despite the documented privacy guarantee; local execution alone does not prevent that disclosure.

Content

No source excerpt is available for this finding.