Back to skill

Security audit

pubmed-verifier

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence, mainly because scanners found citation-verification flaws that could produce misleading results; the review is partial.

Consider the suspicious verdict and do not rely on this skill as the sole check for citation accuracy: reported verification flaws could miss mismatches or misclassify lookup failures. The scanner found external registry requests and reported inadequate disclosure about sending citation data; avoid submitting confidential citation information unless you understand and accept that data flow. Coverage is incomplete because the skillspector scan was partial, including unresolved local references.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (19)

The reference-verification workflow lacks a privacy disclosure about sending citation data to external services.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The quick start instructs users to verify an entire reference list, and line 11 describes OpenAlex title searches, but the README never explicitly warns that citation identifiers or titles are transmitted to third-party services. The 'fully local' description on line 45 can further obscure this data flow, particularly for references associated with unpublished work.

Content

No source excerpt is available for this finding.

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

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 77)May include surrounding context.

md
the input (`--parse-text`/`--stdin`);
- MUST NOT invoke alone: "write me a paper/lit review" (research first —
  cite-holmes; run this skill on the DRAFT's reference list at the end);
- MUST NOT invoke for: plain formatting of UNVERIFIED citations without
  verification intent — formatting stands on verified records
  (`--format-references` verifies first by design);
- Mixed intent ("check the citations in my draft", "帮我润色并核对文献"):

The research workflow sends search keywords to NCBI without a privacy warning.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

Step 1 instructs the agent to transmit research keywords to NCBI's external esearch service, rather than merely listing an available API. The workflow does not warn that user-derived medical research queries leave the local environment or advise excluding sensitive information.

Content

No source excerpt is available for this finding.

The instruction to set `--mailto` lacks a warning that the supplied email address is sent to Crossref.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The speed checklist instructs the agent to set --mailto for Crossref's polite pool but describes only the performance benefit. It does not disclose that this shares the supplied contact email address with a third-party service.

Content

No source excerpt is available for this finding.

Explicit DOI claims can be silently discarded despite the documented promise that explicit DOI inputs are never merged.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The DOI deduplication loop unconditionally skips DOIs already present in PMID results or earlier DOI inputs, without checking whether the skipped input carries explicit claims or different metadata. Consequently, a legitimate PMID or an earlier DOI occurrence can suppress verification of a conflicting explicit DOI claim, omitting its mismatch from reports and potentially allowing a successful exit. This is particularly consequential for a citation-auditing skill whose central purpose is detecting real identifiers paired with fabricated metadata.

Content

No source excerpt is available for this finding.

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · scripts/pv_netio.py (reported line 53)May include surrounding context.

python
parts = url.split("/")
    return "/".join(parts[:3]) if len(parts) > 2 else url

def _api_get(url: str, max_retries: int = 3, timeout=None) -> bytes:
    """HTTP GET hardened for the field: UA rotation on 403/406, Retry-After
    compliance on 429, per-host circuit breaker.

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · scripts/pv_netio.py (reported line 430)May include surrounding context.

python
raise
    return []

def resolve_doi(doi: str, timeout=None) -> dict:
    """Resolve a DOI against the Crossref works API (v2.8.0).

    Returns {'status': 'resolved'|'not_found'|'error', 'meta': dict|None,

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · scripts/pv_netio.py (reported line 498)May include surrounding context.

python
json.JSONDecodeError, ValueError, CircuitOpenError) as e:
        return {"status": "error", "meta": None, "error": str(e)}

def fetch_openalex_doi(doi: str, timeout=None) -> dict:
    """Resolve a DOI against OpenAlex (v4.1.0) — the fallback for DOIs that
    do not live in Crossref (DataCite, Zenodo, SSRN, Chinese registries).

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · scripts/pv_netio.py (reported line 537)May include surrounding context.

python
def search_openalex_title(query: str, max_results: int = 3,
                          raise_on_error: bool = False,
                          timeout=None) -> list:
    """Search OpenAlex works by free-text title (v4.1.0). Raises on network
    failure (callers decide honesty), returns a list of work dicts."""
    params = urllib.parse.quote(query, safe="")

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · scripts/pv_netio.py (reported line 116)May include surrounding context.

python
time.sleep(2 ** attempt)
    if transport:
        _HOST_FAILS[hk] += 1
    if last_err is None:  # e.g. max_retries=0 — never raise None
        raise RuntimeError(f"no attempts made for {url}")
    raise last_err

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 · references/api_examples.md (reported line 97)May include surrounding context.

bash
# ?mailto= joins the polite pool — more generous rate limits
curl -s "https://api.crossref.org/works/10.1038/nature12968?mailto=you%40lab.org"
  • 429 responses carry Retry-After — back off accordingly (v2.2.0 clamps 1–5 s)

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 · scripts/pv_netio.py (reported line 349)May include surrounding context.

python
def fetch_doi_metadata(doi: str) -> dict:
    """Fetch metadata from Crossref API by DOI. With --mailto set, requests
    join the Crossref polite pool (?mailto=) for more generous rate limits."""
    url = f"https://api.crossref.org/works/{urllib.parse.quote(doi, safe='')}"
    if _OPTS["mailto"]:
        url += f"?mailto={urllib.parse.quote(_OPTS['mailto'])}"
    try:

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 · scripts/pv_netio.py (reported line 437)May include surrounding context.

python
'error': str}. A 404 is an authoritative not-found (fabrication signal);
    any transport/other failure is an 'error' and never counts against the DOI.
    """
    url = f"https://api.crossref.org/works/{urllib.parse.quote(_clean_doi(doi), safe='')}"
    if _OPTS["mailto"]:
        url += f"?mailto={urllib.parse.quote(_OPTS['mailto'])}"
    try:

OpenAlex fallback collapses network failures into the same result as an authoritative not-found response.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

fetch_openalex_doi returns the same empty dictionary for an authoritative 404, other HTTP errors, transport failures, and malformed responses. Its documented contract leaves the preceding Crossref-404 interpretation unchanged, so an unavailable OpenAlex fallback can cause legitimate non-Crossref DOIs to be treated as not found rather than unverified; downstream callers cannot recover the lost distinction.

Content

No source excerpt is available for this finding.

CJK verification accepts candidates without year agreement despite documenting a strict year-agreement gate.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The year gate rejects a candidate only when both years exist and differ; missing years pass, and missing claimed metadata is subsequently populated from the candidate itself. An incomplete reference can therefore receive a 'correct' verdict without the documented independent year agreement, weakening the integrity of a tool specifically intended to detect fabricated or misattributed citations.

Content

No source excerpt is available for this finding.

Generic retraction-notice titles are returned as reasons despite an explicit promise not to do so.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

L0306–L0309 explicitly promise that 'Retraction Note: ' alone never masquerades as a reason, reinforced by the function documentation at L0986–L0991. When that generic prefix is stripped and the remainder contains no reason-related terms, L1003 nevertheless returns the complete notice title as the reason, which L0315–L0316 exposes as retraction_reason.

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 · scripts/verify_pmids.py (reported line 1596)May include surrounding context.

python
("Europe PMC", "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=test&format=json&pageSize=1",
     "meta fallback, citation counts, CJK title search",
     "NCBI via --meta-source ncbi"),
    ("Crossref", "https://api.crossref.org/works/10.1038/nature12968",
     "DOI verification and retraction notices",
     "OpenAlex (automatic second lookup)"),
    ("arXiv API", "https://export.arxiv.org/api/query?id_list=1706.03762&max_results=1",

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 · scripts/verify_pmids.py (reported line 1602)May include surrounding context.

python
("arXiv API", "https://export.arxiv.org/api/query?id_list=1706.03762&max_results=1",
     "arXiv preprint verification",
     "none — arXiv-sourced citations will grade unknown"),
    ("OpenAlex (fallback)", "https://api.openalex.org/works?search=test&per-page=1",
     "title search, CJK verification, citation counts",
     ""),
]