Is Claude Blog safe?
- Unsafe deserialization
- Python shell/command execution
- MCP server launches a host command
What to do: Nothing here argues against installing it. Grant the capabilities it lists only if you expect the tool to need them.
Claude Blog is an agent skill analyzed by SkillTotal's deterministic static scanner. The scan found no malicious indicators, though 1 risky construct is reported for review. It can: delegated authentication, filesystem read, filesystem write, mcp tools detected, network egress and shell execution — capabilities are what the code can do, not a verdict on intent. Risk score 20/100 (low).
claude-blog 2.3.2
Get notified when this component changes
Change alerts aren't live yet. Leave your email and we'll let you know when they launch. We use it only for this and delete it on request.
Automated static-analysis result. It can contain false positives and false negatives, and is not a claim about the intent of Claude Blog's authors. Report a false positive.
Behavioral traits
How this component maps to the CSA agentic threat model. Descriptive — it never affects the risk score.
Findings (8)
It loads data with a format that can rebuild arbitrary objects (e.g. pickle, or unsafe YAML).
value = yaml.load("\n".join(lines[1:end]), Loader=UniqueKeyLoader)Why it matters: Feeding such a loader untrusted data can execute code hidden inside that data.
Fix: Deserialize untrusted data with a safe format/loader: JSON, or yaml.safe_load / Loader=SafeLoader. Reserve pickle/marshal for data you fully control.
An MCP server entry launches a command on your host.
"command": "npx",
Why it matters: Trusting the manifest means running that binary — verify what it is and where it comes from.
Fix: Verify the launched command and its source before trusting this MCP server configuration.
The component can run operating-system commands or spawn processes.
result = subprocess.run(
[args.cli, "plugin", "validate", args.root],
capture_output=True,
text=True,
check=False,
)return subprocess.call([PY, str(REPO / "scripts" / script), *args], cwd=REPO)
return subprocess.run([PY, str(REPO / "scripts" / script), *args], cwd=REPO, text=True, capture_output=True, check=False)
return subprocess.run(
command,
cwd=REPO,
text=True,
capture_output=True,
timeout=EXEC_TIMEOUT_SECONDS,
check=False,
)subprocess.run([PY, *args], cwd=REPO, check=True)
proc = subprocess.run(["git", "diff", "--quiet", "--", rel], cwd=REPO, check=False)
proc = subprocess.run(
[sys.executable, str(audit), "--require", "market-ready", "--json"],
cwd=REPO,
text=True,
capture_output=True,
check=False,
)proc = subprocess.run(["git", "ls-files", "-z", "--cached"], cwd=REPO, capture_output=True, check=False)
proc = subprocess.run(args, cwd=REPO, capture_output=True, check=False)
proc = subprocess.run(["git", "ls-files", "-z", "--others", "--exclude-standard"], cwd=REPO, capture_output=True, check=False)
proc = subprocess.run(["git", "rev-parse", "HEAD"], cwd=REPO, capture_output=True, text=True, check=False)
result = subprocess.run(
["pdftotext", "-layout", pdf_file.name, text_file.name],
text=True,
capture_output=True,
check=False,
timeout=30, …result = subprocess.run(
command,
cwd=ROOT,
text=True,
capture_output=True,
check=False,
)result = subprocess.run(
["git", *args], check=False, capture_output=True
)result = subprocess.run(
["gh", "auth", "token"], capture_output=True, text=True
)subprocess.run(
["ffmpeg", "-y", "-i", wav_path, "-codec:a", "libmp3lame",
"-b:a", "192k", "-ar", "24000", "-ac", "1", str(tmp_path)],
check=True, capture_output=True,
)result = subprocess.run(cmd)
subprocess.run(
[str(self.venv_pip)] + install_args,
check=True, capture_output=True, text=True,
timeout=PIP_TIMEOUT_SECONDS,
)result = subprocess.run(cmd)
subprocess.run(
[str(self.venv_pip)] + install_args,
check=True, capture_output=True, text=True,
timeout=PIP_TIMEOUT_SECONDS,
)result = subprocess.run(
["git", "ls-files", "--error-unmatch", path.name],
cwd=path.parent,
capture_output=True,
text=True,
)result = subprocess.run(cmd)
result = subprocess.run(
[str(self.venv_pip)] + install_args,
check=True,
capture_output=True,
text=True,
timeout=PIP_TIMEOUT_SECONDS, …subprocess.run(
[str(self.venv_python), "-m", "patchright", "install", "chrome"],
check=True,
capture_output=True,
text=True, …result = subprocess.run(cmd)
Why it matters: Powerful and often legitimate — confirm the commands aren't built from untrusted input.
Fix: Confirm the command and its arguments are fully controlled and not derived from untrusted input; avoid shell=True.
The component reads files from disk.
text = Path(args.from_file).read_text(encoding="utf-8")
text = path.read_text(encoding="utf-8", errors="replace")
text = path.read_text(encoding="utf-8", errors="replace")
frontmatter, _body = split_frontmatter(path.read_text(encoding="utf-8", errors="replace"))
text = path.read_text(encoding="utf-8", errors="replace")
text = path.read_text(encoding="utf-8", errors="replace")
text = path.read_text(encoding="utf-8", errors="replace")
data = json.loads(path.read_text(encoding="utf-8"))
return path.read_text(encoding="utf-8", errors="replace")
marked = json.loads(demo_marker.read_text(encoding="utf-8"))
data = json.loads(marker.read_text(encoding="utf-8"))
links += len(re.findall(r"!?\[\[([^\]]+)\]\]", path.read_text(encoding="utf-8", errors="replace")))
data = json.loads(manifest_path.read_text(encoding="utf-8"))
text = hot.read_text(encoding="utf-8") if hot.exists() else ""
data = json.loads(Path(path).read_text(encoding="utf-8"))
schema = json.loads(SCHEMA_PATH.read_text(encoding="utf-8"))
raw_bytes = source.read_bytes()
data = json.loads(Path(path).read_text(encoding="utf-8"))
schema = json.loads(SCHEMA_PATH.read_text(encoding="utf-8"))
raw = source.read_bytes()
data = path.read_bytes()
data = json.loads(path.read_text(encoding="utf-8"))
text = log.read_text(encoding="utf-8").rstrip() + f"\n- {date.today().isoformat()} - {message}\n"data = json.loads(Path(path).read_text(encoding="utf-8"))
schema = json.loads(SCHEMA_PATH.read_text(encoding="utf-8"))
Why it matters: Usually legitimate, but worth confirming it can't be steered into reading sensitive files.
Fix: Confirm which files are read and that paths cannot be influenced by untrusted input to reach sensitive locations.
The component writes or deletes files on disk.
(temp_demo / DEMO_MARKER).write_text(DEMO_MARKER_TEXT, encoding="utf-8")
shutil.rmtree(demo)
shutil.copytree(temp_demo, demo)
Path(output).write_text(text, encoding="utf-8")
Path(output).write_text(text, encoding="utf-8")
shutil.copy2(source, target)
source_note.write_text(f"""---
type: {yaml_string("source")}
title: {yaml_string(display_title)}
domain: {yaml_string("claude-blog-brain")}
created: {yaml_string(date.today().isoformat())}
updated: {yaml_string(date.today().isoformat())}
st …path.write_text(json.dumps(data, indent=2) + "\n", encoding="utf-8")
log.write_text(text, encoding="utf-8")
Path(output).write_text(text, encoding="utf-8")
manifest_path.write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n", encoding="utf-8")
path.write_text("\n".join(f"{sha256_file(a)} {a.name}" for a in artifacts) + "\n", encoding="utf-8")Path(output).write_text(markdown, encoding="utf-8")
Path(output).write_text(markdown, encoding="utf-8")
Path(output).write_text(markdown, encoding="utf-8")
manifest.write_text(json.dumps({
"scaffold_history": [{
"client": slug,
"client_name": args.client_name or slug,
"owner": args.owner,
"created": date.today().isofor …shutil.rmtree(vault)
shutil.copytree(temp_vault, vault)
target.write_bytes(data)
target.write_text(text, encoding="utf-8")
Path(output).write_text(text, encoding="utf-8")
log.write_text(log.read_text(encoding="utf-8").rstrip() + f"\n- {date.today().isoformat()} - {message}\n", encoding="utf-8")LEDGER_PATH.write_text(json.dumps(ledger, indent=2, ensure_ascii=False) + "\n", encoding="utf-8")
analyze_blog._safe_write_text(output_path, output)
os.unlink(tmp)
Why it matters: Usually legitimate, but worth confirming the paths can't be controlled by untrusted input.
Fix: Confirm which files are written/deleted and that paths cannot be influenced by untrusted input.
The component makes outbound network requests.
import urllib.request
parsed = urllib.parse.urlparse(ref)
raw_path = urllib.parse.unquote(parsed.path)
parsed = urllib.parse.urlsplit(url.strip())
return urllib.parse.urlunsplit(
(parsed.scheme.lower(), parsed.netloc.lower(), parsed.path or "/", parsed.query, "")
)parsed = urllib.parse.urlparse(url)
parsed = urllib.parse.urlparse(url)
if urllib.parse.urlparse(url).scheme != "https":
parsed = urllib.parse.urlparse(url)
name = urllib.parse.unquote(urllib.parse.urlsplit(src).path).rsplit("/", 1)[-1]parsed = urllib.parse.urlparse(route.request.url)
req_path = Path(urllib.request.url2pathname(parsed.path)).resolve()
if canonical and not urllib.parse.urlsplit(value).scheme:
urllib.parse.urljoin(canonical, value),
urllib.parse.urljoin(canonical.rstrip("/") + "/", value),parsed = urllib.parse.urlsplit(candidate)
path = urllib.parse.quote(urllib.parse.unquote(parsed.path), safe="/:@-._~!$&'()*+,;=")
result.add(urllib.parse.urlunsplit((parsed.scheme.lower(), host, path, parsed.query, "")))
_PREFLIGHT_NO_REDIRECT_OPENER = urllib.request.build_opener(urllib.request.ProxyHandler({}), _PreflightNoRedirectHandler())req = urllib.request.Request(url, method=method, headers=request_headers)
redirect_url = urllib.parse.urljoin(url, location)
literal = ipaddress.ip_address(urllib.parse.urlparse(href).hostname or "")
target = urllib.parse.unquote(href[1:])
import urllib.request
parsed = urllib.parse.urlparse(value)
Why it matters: Usually legitimate, but confirm the destinations are expected and no sensitive data leaves.
Fix: Confirm the destination hosts are expected and that no sensitive data is sent off-host.
An OAuth 2.0 / OpenID Connect delegated-authentication flow was detected (authorization-code / refresh-token / token-exchange grant, an OIDC authorize/discovery endpoint or id_token, or a delegation library). Tools authenticate with the end user's delegated, scoped credentials rather than a long-lived embedded service credential. (3 occurrence(s) shown as evidence).
r"api[_-]?key|key|access_token|refresh_token|id_token|client_secret|"
"grant_type": "refresh_token",
"grant_type": "authorization_code",
Fix: Delegated auth is a lower-blast-radius execution context than an embedded static credential. Confirm the requested scopes are minimal and that tokens are never logged or forwarded off-host.
An MCP tool surface (manifest or tool definitions) was found.
"mcpServers": {Why it matters: Just context — review which tools it offers and their permissions.
Fix: Review the declared MCP tools and their permissions.
How attackers abuse these capabilities
Interactive labs on the attack class behind the rules above. They show the technique, not anything found in Claude Blog.
Maintain Claude Blog?
The free Claude Code plugin scans each npm, PyPI or MCP package with this report's open-source engine before the install command runs. It runs on your machine and needs no account.
Check your own component
Run the same evidence-backed scan on any MCP server, agent skill, or package.
Scan your own componentHow we determine this: deterministic static analysis (regex + AST), evidence-anchored, no code execution. Methodology →