Back to skill

Security audit

job-hunt

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with high confidence because scanners report broad, undisclosed capabilities, excessive access and execution risks, and incomplete analysis; this is a review warning, not a finding that the author acted maliciously.

Consider the verdict a reason to review carefully before installing. The scanner reports recurring scheduling, broad browser access, host-level command execution, and some credential or external-service use, while parts of the artifact were only partially analyzed. The results do not establish malicious intent, but they also do not establish that the artifact is safe.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
Findings (124)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The digest's job-search content aligns with the declared discover mode, and local state and deduplication are supporting implementation details. However, installing persistent OS scheduler entries, initiating recurring agent sessions, and orchestrating an email-based subscription workflow are material capabilities absent from the description. These operations are explicitly exposed through setup, install, tick, inbox, and delivery commands; they are not merely incidental document-generation mechanics. The code includes pause/cancel controls and a macOS confirmation dialog, but those safeguards do not resolve the disclosure mismatch. There is no evidence in this chunk of direct email transmission or unrelated data exfiltration.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied code's purpose is maintaining and advertising updates for OpenCLI, rather than performing the declared job-assistance tasks. Although update checking can be supporting infrastructure, the supplied description establishes no connection to OpenCLI or its extension, making these tool-maintenance behaviors unrelated to the declared purpose. The code does not automatically install updates, and no transfer of resumes or other user documents is shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied script's primary function is source-code security and repository maintenance, not any declared job-search or application workflow. Although it could serve as a development-time quality gate for the skill, it is a standalone repository audit rather than an implementation step that produces or validates the declared career-assistance outputs. The shown code does not modify scanned files or transmit their contents.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
89% confidence
Finding

Updating job-search preferences from role feedback supports the declared discovery mode. However, managing a recurring email digest and its subscription lifecycle is an additional user-facing capability not disclosed in the description. The chunk processes messages supplied to it; it does not itself demonstrate mailbox access or email sending. The mismatch concerns the undeclared digest-management functionality, not ordinary file persistence or the absence of other advertised job-assistance features from this individual chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
87% confidence
Finding

The reported cookie-value retrieval, unrestricted response capture, and arbitrary CDP/application control exceed the access needed to read job postings. Reusing the user's daily authenticated browser increases the potential exposure to session theft, unrelated private data, and unauthorized actions; however, the underlying implementations are not included in this source, so their reachability remains unverified.

Content

No source excerpt is available for this finding.

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 77)May include surrounding context.

md
[references/portability.md](references/portability.md); before rendering a CV,

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 204)May include surrounding context.

md
Run `scripts/deliver.py --workspace <ws>` last. It copies client files to

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 216)May include surrounding context.

md
named on `deliver.py`'s last line. Keep audit records private. In discover,

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 219)May include surrounding context.

md
If `deliver.py` refuses, keep the draft in the workspace and say what is missing;

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 222)May include surrounding context.

md
`deliver.py`'s output rather than retyping them.

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · agents/ats-screener.md (reported line 79)May include surrounding context.

md
Location does not change the mechanical count. Flag any required term that appears only in prose but would benefit from also appearing in a dedicated Skills section.

**Skills-only signal note:** if a must-have is `present` *only* because it appears in the Skills section with no corresponding mention anywhere in experience/prose, mark it `[Skills-only]`. It still counts as present for coverage — but flag it so the human reviewer can assess whether it is genuinely substantiated. You do not judge substantiation; you only report the location asymmetry. (This is the clean hand-off: you report skills-only matches, the Hiring Manager judges whether they are real.)

---

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · references/workflow-checklist.md (reported line 215)May include surrounding context.

md
Ran, leaving a `mode_entry` record rather than a gate receipt:
- [ ] `scripts/enter_mode.py` — and its recorded hash still matches `modes/apply.md`.

Ran, leaving nothing in the journal (they render; they do not judge):
- [ ] `scripts/render_cv.py`, plus `scripts/render_letter.py` /
      `scripts/render_rirekisho.py` if applicable.
- [ ] `scripts/deliver.py` — the LAST step of every mode. A workspace under

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/opencli-cdp/cdp.js (reported line 335)May include surrounding context.

js
async startNetworkCapture(pattern = '') {
        // Always update the filter pattern
        this._networkCapturePattern = pattern;
        // Reset state only on first start; avoid wiping entries if already capturing
        if (!this._networkCapturing) {
            this._networkEntries = [];
            this._pendingRequests.clear();

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · references/market-conventions/nl.yaml (reported line 387)May include surrounding context.

yaml
why: |-
      Postings never describe the screening step. Candidates either volunteer referees far too
      early, or assume — from markets where quiet back-channel checks are routine — that a current
      employer may be contacted without warning, and so avoid applying while employed.
  - id: "nl-pay-range-not-pay-history"
    text_en: |-
      Pay history is not supposed to be the anchor here. The NVP Sollicitatiecode states as a founding

External Model or Provider Selection

High
Category
Excessive Agency
Confidence
90% confidence
Finding

Skill selects an external model or provider that may use a different account or billing plan than the operator expects. Undisclosed model switches can cause unexpected cost or quota consumption.

Content

Scanner excerpt · references/portability.md (reported line 65)May include surrounding context.

portable. The persona files are standalone, so:

bash
codex exec --skip-git-repo-check --ephemeral -s read-only "$(cat agents/ats-screener.md)

POSTING:
$(cat "$WS/posting.yaml")

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · scripts/consistency.py (reported line 91)May include surrounding context.

python
"""How many requirements were marked hard filters, when that count is high enough
    to mean the rule was applied loosely.

    A knockout is a condition a recruiter can check without judgement and which ends
    the application on its own. Postings label half their list "required", and the whole
    point of the screening axis is to be the smaller, harder list underneath that label.
    When it stops being smaller it has collapsed back into the thing it was separating

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/lint_no_prediction.py (reported line 1203)May include surrounding context.

python
`stdout_file` is stored workspace-relative by the wrapper. A path that escapes
    the workspace is dropped rather than followed: a journal is data, and a data
    file that can name `/etc/passwd` as a capture is a traversal, not a corpus.
    """
    root = workspace.resolve()
    out, seen = [], set()

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 · scripts/mutants.py (reported line 210)May include surrounding context.

python
# pristine .pyc and "survives" without ever running, so a baseline recorded
# before this fix may list survivors that were never tested. Children inherit
# the env, so the python subprocesses the suite spawns write no cache either.
_ENV = {**os.environ, "PYTHONDONTWRITEBYTECODE": "1"}


def _run_pytest(cwd: pathlib.Path, targets: list) -> subprocess.CompletedProcess:

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
93% confidence
Finding

This is the same unconfined execution boundary identified by the AST4 finding: cmd and build are executed as shell commands with host privileges. Explicit command parameters are expected for this tool, but there is no visible containment protecting the user's job-search documents and credentials from hostile candidate code or commands.

Content

Scanner excerpt · scripts/run_mock_tests.py (reported line 117)May include surrounding context.

python
ILL)
        except (ProcessLookupError, PermissionError):
            pass


def _run(cmd: str, qdir: pathlib.Path, stdin: str, timeout: float) -> dict:
    started = datetime.datetime.now()
    # Its own process group, so a timeout reaches the solution and not just the shell.
    # UTF-8 both ways: a locale codec (cp936 on a Chinese Windows) misreads the output.
    group = ({"creationflags": subprocess.CREATE_NEW_PROCESS_GROUP} if os.name == "nt"
             else {"start_new_session": True})
    proc = subprocess.Popen(cmd, shell=True, cwd=qdir, stdin=subprocess.PIPE,
                            stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True,
                            encoding="utf-8", errors="replace", **group)
    try:
        out, err = proc.communicate(stdin, timeout=timeout)
    except subprocess.TimeoutExpired:
        _kill_tree(proc)
        try:
            out, _ = proc.communicate(timeout=5)
        except subprocess.TimeoutExpired:
            out = ""
        return {"timed_out":

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 · scripts/run_tool.py (reported line 36)May include surrounding context.

python
return 2
    budget_file = None
    runtime = json.loads((args.root.expanduser() / "runtime.json").read_text())
    env = os.environ.copy()
    paths = [str(Path(runtime["python"]).parent)]
    if "node" in runtime:
        paths += [str(Path(runtime["node"]).parent), str(Path(runtime["opencli"]).parents[4] / ".bin")]

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 · scripts/setup_dependencies.py (reported line 169)May include surrounding context.

python
prefix = root / "opencli"
    package = prefix / "node_modules/@jackwener/opencli"
    if not (package / "package.json").exists():
        env = dict(os.environ, PATH=str(node.parent) + os.pathsep + os.environ.get("PATH", ""))
        run([node, npm_entry(node), "install", "--prefix", prefix, "@jackwener/opencli@1.8.7",
             "--ignore-scripts", "--no-audit", "--no-fund"], env=env)
    patch_opencli(package)

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · third_party/anysearch/anysearch_cli.js (reported line 22)May include surrounding context.

js
// END GENERATED:CONSTANTS

function loadEnv() {
  const envPaths = [path.join(__dirname, ".env"), path.join(__dirname, "..", ".env")];
  for (const envPath of envPaths) {
    if (fs.existsSync(envPath)) {
      const lines = fs.readFileSync(envPath, "utf-8").split(/\r?\n/);

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · third_party/anysearch/anysearch_cli.js (reported line 22)May include surrounding context.

js
// END GENERATED:CONSTANTS

function loadEnv() {
  const envPaths = [path.join(__dirname, ".env"), path.join(__dirname, "..", ".env")];
  for (const envPath of envPaths) {
    if (fs.existsSync(envPath)) {
      const lines = fs.readFileSync(envPath, "utf-8").split(/\r?\n/);

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · third_party/anysearch/anysearch_cli.js (reported line 37)May include surrounding context.

js
// Strip surrounding quotes (any number, either kind) and re-trim, to
        // match the Python reference.
        const val = line.substring(idx + 1).trim().replace(/^["']+/, "").replace(/["']+$/, "").trim();
        // Skip empty values so an empty .env entry does not clobber a real
        // environment variable.
        if (key && val) process.env[key] = val;
      }

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · third_party/anysearch/shared/doc_spec.md (reported line 236)May include surrounding context.

md
---

## Rate Limit Handling
- On rate limit error with auto_registered api_key in response: present key to user for approval, then save to .env and retry
- On anonymous quota exhausted: inform user that a key provides higher limits; suggest configuring one via .env or environment variable