ClawHub
Back to skill

Security audit

Job Application

Security checks for vulnerabilities and agentic risk

Overview

This skill locally creates job-application documents with disclosed private-data handling and no evidence of hidden submission, exfiltration, persistence, or privilege escalation.

Use this only if you are comfortable giving the skill access to your local profile data and running local PDF tools. Configure JOB_APPLICATION_PROFILE_DIR and output/data directories to private locations, keep validation enabled for final documents, review generated files before attaching them, and authorize any chat delivery or application submission one destination at a time.

A.I.G

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

SkillSpector

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (28)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
70% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · README.en.md (reported line 74)May include surrounding context.

md
`render_application.py` runs `validate_application.py` by default. It requires
readable CV and cover-letter PDFs, exactly one CV page, no unresolved
placeholders, valid input schemas, in-page text bounds, and useful CV vertical
fill. `--skip-validation` is a visible debug-only escape hatch.

Use `--strict-fill` to make an underfilled CV fail. `--min-vertical-fill` and
`--max-vertical-fill` default to `0.52` and `0.98`.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
70% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · README.md (reported line 95)May include surrounding context.

md
`render_application.py` runs `validate_application.py` by default. It requires
readable CV and cover-letter PDFs, exactly one CV page, no unresolved
placeholders, valid input schemas, in-page text bounds, and useful CV vertical
fill. `--skip-validation` is a visible debug-only escape hatch.

Use `--strict-fill` to make an underfilled CV fail. `--min-vertical-fill` and
`--max-vertical-fill` default to `0.52` and `0.98`.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The declared description presents a content-generation tool for drafting multiple job-application artifacts, potentially including PDF rendering. The actual code shown only parses command-line arguments, loads existing JSON files, converts legacy cover-letter and/or plain-text application data into a newer schema using helper functions, and writes the converted JSON output. It does not create a CV, does not submit applications, and does not perform PDF rendering. While the domain is related to job-application materials, the code's primary purpose is format conversion of existing content, which is materially different from the declared drafting/rendering functionality.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared purpose is about generating job-application materials, but the actual code chunk does not generate, tailor, validate, or render application content. Instead, it performs filesystem packaging: it whitelists specific files, checks for symlinks/unsafe paths, enforces an external empty output directory, and copies public assets into that destination. This is a materially different primary purpose from drafting/rendering application documents. While the copied files belong to the same skill, this code's behavior is an internal publishing/export tool, an undeclared capability relative to the stated description.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The code’s actual function is narrowly scoped to selecting relevant CV record IDs from profile collections based on keyword overlap with job data and configured limits. It loads profile/job JSON, computes ranked selections, optionally derives from an existing base selection, validates the result, and writes a JSON output. There is no functionality for composing tailored prose documents, generating cover letters, producing a plain-text application, or rendering PDFs. This is a material description-behavior mismatch because the declared purpose describes end-to-end application-material creation, while the supplied code chunk only builds an intermediate selection artifact for CV content.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The code chunk does not implement the declared core functionality of generating tailored job-application materials. Instead, it is a utility that prepares an application folder, optionally validates an existing cv_selection.json from a base application, and creates or reuses a directory on disk. While this may support a larger application-generation workflow, the actual behavior in this chunk is materially different from the declared primary purpose. No evidence of submission behavior appears, but the mismatch is still significant because the described outputs and capabilities are absent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The declared description says the skill creates job-application materials and optionally renders PDFs. The supplied code does not generate CVs, cover letters, or plain-text applications, nor does it render PDFs. Instead, its primary function is post-generation validation of local output files. It reads local PDFs and JSON/text files, runs Poppler utilities, parses PDF text and coordinates, enforces one-page CV and fill constraints, detects placeholders, and confirms consistency with profile/application inputs. This is a materially different primary purpose from content creation/rendering, so the description does not accurately represent the code chunk.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
84% confidence
Finding

Documenting a visible --skip-validation switch creates a straightforward path to bypass the integrity and consistency checks that are intended to prevent bad or mismatched application artifacts from being distributed. In this skill's context, those checks protect against leaking fictional/bundled profile data, malformed outputs, or unreviewed content being attached or shared externally.

Content

Scanner excerpt · SKILL.md (reported line 63)May include surrounding context.

python3 scripts/render_application.py

text

Rendering runs `validate_application.py` by default; Poppler's `pdfinfo` and `pdftotext` are required. The visible `--skip-validation` switch is for debugging only. Use `validate_application.py <application-dir> --strict-fill` to make an underfilled CV fail. For an external profile, pass the same `--profile-dir` to the cover-letter generator and renderer (or configure `JOB_APPLICATION_PROFILE_DIR`). The generator refuses an external letter that would silently use the bundled fictional profile, and the validator checks the letterhead against the selected profile.
On Windows, `py -3` can be used instead of `python3`. Linux and macOS users may alternatively use the `scripts/render_application.sh` wrapper.
7. Return the generated paths and validation result. Deliver files externally only after destination-specific approval.
- If the user explicitly asks to receive the documents in the current chat and the channel supports attachments, send the validated `CV.pdf`, `cover_letter.pdf`, and (when useful) `indeed_text.txt` as native downloadable document attachments in that conversation.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

The --skip-validation flag allows callers to bypass the mandatory post-render checks that are meant to detect malformed or incomplete output. In an agent setting, a controllable bypass parameter can be abused to suppress safeguards and permit unsafe or low-integrity artifacts to be treated as successful output.

Content

Scanner excerpt · scripts/render_application.py (reported line 46)May include surrounding context.

python
parser.add_argument("application_directory", type=Path)
    parser.add_argument("--profile-dir", type=Path, default=PROFILE_DIR,
                        help="Profile directory used by post-render validation")
    parser.add_argument("--skip-validation", action="store_true",
                        help="DEBUG ONLY: do not run the mandatory post-render quality checks")
    parser.add_argument("--strict-fill", action="store_true",
                        help="Treat a too-empty CV as a rendering failure")

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill declares shell, file read/write, and environment-variable-driven behavior but does not define an explicit tool/permission scope. That creates an avoidable trust gap: an agent invoking this skill may grant broader capabilities than users expect, including access to private profile data and local filesystem locations used for application generation and export.

Content

No source excerpt is available for this finding.

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
81% confidence
Finding

Even though validation runs by default, exposing a skip-validation path in the documented workflow is an unsafe default-adjacent design because agents or users may copy the command and disable safeguards for convenience. In a skill that handles private profile data and externally shareable application documents, bypassing validation increases the chance of privacy leaks, inconsistent identity data, or distribution of incomplete artifacts.

Content

Scanner excerpt · SKILL.md (reported line 63)May include surrounding context.

python3 scripts/render_application.py

text

Rendering runs `validate_application.py` by default; Poppler's `pdfinfo` and `pdftotext` are required. The visible `--skip-validation` switch is for debugging only. Use `validate_application.py <application-dir> --strict-fill` to make an underfilled CV fail. For an external profile, pass the same `--profile-dir` to the cover-letter generator and renderer (or configure `JOB_APPLICATION_PROFILE_DIR`). The generator refuses an external letter that would silently use the bundled fictional profile, and the validator checks the letterhead against the selected profile.
On Windows, `py -3` can be used instead of `python3`. Linux and macOS users may alternatively use the `scripts/render_application.sh` wrapper.
7. Return the generated paths and validation result. Deliver files externally only after destination-specific approval.
- If the user explicitly asks to receive the documents in the current chat and the channel supports attachments, send the validated `CV.pdf`, `cover_letter.pdf`, and (when useful) `indeed_text.txt` as native downloadable document attachments in that conversation.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The legacy content conversion logic injects fixed German phrases such as "Guten Tag," and "Mit freundlichen Grüßen" when values are missing. This forces a specific locale in generated content without any visible user opt-in or language choice at these call sites, which fits the language/locale policy violation category.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This legacy plain-text conversion path populates multiple fallback strings in German, including "Nicht angegeben", "Guten Tag,", and "Mit freundlichen Grüßen". These hard-coded defaults impose a specific language/locale without explicit user choice in this code path.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
92% confidence
Finding

The script invokes an external LaTeX engine on user-selected .tex files, and TeX is not a passive data format: depending on engine configuration and document contents, compilation can trigger dangerous file reads, shell escapes, or resource-exhaustion behavior. Because this skill renders application materials derived from structured input, untrusted content flowing into LaTeX makes subprocess execution materially riskier than a normal fixed-command helper.

Content

Scanner excerpt · scripts/render_application.py (reported line 26)May include surrounding context.

python
]
    outputs = []
    for pass_number in (1, 2):
        completed = subprocess.run(
            command,
            cwd=source.parent,
            check=False,

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

This skill expands from document drafting into execution of an external renderer, which creates an attack surface outside normal text generation. In this context, rendering attacker-influenced LaTeX can expose the host to command execution features, filesystem access, or denial-of-service through the TeX toolchain.

Content

No source excerpt is available for this finding.

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · README.en.md (reported line 74)May include surrounding context.

md
parser.add_argument("application_directory", type=Path)
    parser.add_argument("--profile-dir", type=Path, default=PROFILE_DIR,
                        help="Profile directory used by post-render validation")
    parser.add_argument("--skip-validation", action="store_true",
                        help="DEBUG ONLY: do not run the mandatory post-render quality checks")
    parser.add_argument("--strict-fill", action="store_true",
                        help="Treat a too-empty CV as a rendering failure")

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · README.md (reported line 95)May include surrounding context.

md
parser.add_argument("application_directory", type=Path)
    parser.add_argument("--profile-dir", type=Path, default=PROFILE_DIR,
                        help="Profile directory used by post-render validation")
    parser.add_argument("--skip-validation", action="store_true",
                        help="DEBUG ONLY: do not run the mandatory post-render quality checks")
    parser.add_argument("--strict-fill", action="store_true",
                        help="Treat a too-empty CV as a rendering failure")

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · scripts/render_application.py (reported line 46)May include surrounding context.

python
parser.add_argument("application_directory", type=Path)
    parser.add_argument("--profile-dir", type=Path, default=PROFILE_DIR,
                        help="Profile directory used by post-render validation")
    parser.add_argument("--skip-validation", action="store_true",
                        help="DEBUG ONLY: do not run the mandatory post-render quality checks")
    parser.add_argument("--strict-fill", action="store_true",
                        help="Treat a too-empty CV as a rendering failure")

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · scripts/render_application.py (reported line 75)May include surrounding context.

python
(app_dir / "build.log").write_text("\n".join(logs), encoding="utf-8")

    print(f"Rendered {app_dir / 'CV.pdf'} and {app_dir / 'cover_letter.pdf'}")
    if args.skip_validation:
        print("WARNING: post-render application validation was skipped (debug option).")
        return 0
    report = validate_application(

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/render_example.py (reported line 18)May include surrounding context.

python
def run(script: str, *arguments: Path | str) -> None:
    subprocess.run(
        [sys.executable, str(SCRIPTS_DIR / script), *(str(argument) for argument in arguments)],
        check=True,
    )

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/validate_application.py (reported line 57)May include surrounding context.

python
def _run(command: list[str]) -> str:
    completed = subprocess.run(command, capture_output=True, text=True, encoding="utf-8", errors="replace")
    if completed.returncode:
        detail = (completed.stderr or completed.stdout).strip()
        raise PdfInspectionError(f"{' '.join(command[:2])} failed: {detail or 'unknown error'}")

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 · tests/test_v2.py (reported line 32)May include surrounding context.

python
def run_script(name: str, *args: str) -> subprocess.CompletedProcess[str]:
    return subprocess.run(
        [sys.executable, str(SCRIPTS_DIR / name), *args],
        cwd=SKILL_DIR,
        text=True,

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest-like JSON sets the job language to "de" with no indication that the user can choose another language or that the skill is restricted to German-language applications. This is a natural-language locale constraint and may violate policy unless the locale limitation is explicitly optional or justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

This JSON content is entirely hard-coded in German, including the company details, salutation, body paragraphs, and closing. Because the skill file provides no indication that German is optional or required for a justified region-specific purpose, it presents a natural-language locale constraint without user opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The file content is entirely in German, which indicates a language-specific output choice. Because there is no accompanying natural-language indication that the user opted into German or that the skill is explicitly region-specific, this may violate the language/locale policy requiring user choice or documented justification.

Content

No source excerpt is available for this finding.