Back to skill

Security audit

Notebooklm Ingest

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence, citing excessive agency concerns and incomplete inspection of much of the artifact.

Consider the reported authenticated-request and scheduling concerns before installing. Inspection was partial: the scanner reports 8 of 22 components scanned, with static analysis degraded for some files, so the available evidence is incomplete. The verdict is suspicious, not a finding that the author acted maliciously.

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
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (28)

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 57)May include surrounding context.

md
node scripts/notebooklm-ingest.js notebooks

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 58)May include surrounding context.

md
node scripts/notebooklm-ingest.js notebooks --set "<notebook-id>,<notebook-id>"

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 59)May include surrounding context.

md
node scripts/notebooklm-ingest.js notes --notebook <notebook-id>

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 61)May include surrounding context.

md
| node scripts/notebooklm-ingest.js fetch

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 62)May include surrounding context.

md
node scripts/notebooklm-ingest.js push --dry-run

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 63)May include surrounding context.

md
node scripts/notebooklm-ingest.js push

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 64)May include surrounding context.

md
echo '{"headline":"two notes into your wiki"}' | node scripts/notebooklm-ingest.js report

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 65)May include surrounding context.

md
node scripts/notebooklm-ingest.js history --limit 20

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 66)May include surrounding context.

md
node scripts/notebooklm-ingest.js schedule --cron "0 7 * * *" --tz America/Los_Angeles

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 67)May include surrounding context.

md
node scripts/notebooklm-ingest.js schedule --off

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 68)May include surrounding context.

md
node scripts/notebooklm-ingest.js host --set javis-mac

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 69)May include surrounding context.

md
node scripts/notebooklm-ingest.js doctor

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 342)May include surrounding context.

md
- **Normalise, then hash.** `scripts/normalize.js` is the only place that

YARA rule 'info_stealer': Information stealer patterns (credential harvesting, browser data theft) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · references/credential-contract.md (reported line 38)May include surrounding context.

md
he research spike, with no
automation challenge (F49). In HiJavis the same thing happens through the in-app
sign-in webview, the mechanism Luma's cookie capture already uses.

**Javis never types a Google credential.** Not into a form, not into a CLI
prompt, not from a stored value. If a path seems to require it, that path is
wrong.

Chrome's own cookie store is **not** an alternative: `--browser-cookies chrome`
failed on every profile with "macOS Keychain item not found" (F51), and a
broader sweep of other browsers' cookie stores was deliberately not performed.
Reading a browser's credential store is not a fallback this design has.

## What this bundle may say about it

**Path, presence, size, mtime, and a sha256 digest of the file.** That is the
whole of `credentialFingerprint()`, and it is what `doctor` reports.

**Never a cookie name, and never a cookie value.** Not in stdout, not in a
digest pushed to the user's chat, not in a log line, not in a refusal's detail
field, and never t

YARA rule 'agent_skill_prompt_injection_hidden_instructions': Prompt injection or hidden instructions embedded in AI agent skill text [agent_skills]

High
Category
YARA Match
Confidence
80% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · scripts/notebooklm-ingest.js (reported line 204)May include surrounding context.

js
verything below is gmail-wiki-ingest's
 * sanitizer (gmail-wiki-ingest.js:354-444), carried across wholesale rather
 * than re-derived, so that a hostile title is **visible but inert**: it renders
 * as literal text, it cannot forge a bullet or a footer line, and nothing
 * downstream feeds it back to the agent as instruction.
 *
 * The agent's own headline goes through the identical treatment: a model that
 * has just read a hostile title is perfectly capable of relaying one.
 */
const MAX_TITLE_CHARS = 80;
const MAX_PROSE_CHARS = 120;

const URL_RE = /(?:[a-z][a-z0-9+.-]*:\/\/|www\.|mailto:|data:|javascript:)\S*/gi;
const EMAIL_RE =
  /([^\s@<>()[\]",;:]+)@([a-z0-9](?:[a-z0-9-]*[a-z0-9])?(?:\.[a-z0-9](?:[a-z0-9-]*[a-z0-9])?)+)/gi;
// The backslash is FIRST in the class: a title that supplies its own `\` before
// an active character would otherwise consume the escape we are about to add.
const MD_ACTIVE_RE = /[\\*_[\]()`<>]/g;
const LINE_BREAKING_RE = new RegExp('[\\u0000-\\u001F\\u0

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
100% 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 · test/cron.test.js (reported line 68)May include surrounding context.

js
});

test('R13: a hostile schedule string stays one argv element', () => {
  const argv = cron.buildCronAdd({ cron: '0 7 * * * ; rm -rf /' });
  assert.equal(argv.filter((a) => a.includes('rm -rf')).length, 1);
});

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
100% 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 · test/nlm-cli.test.js (reported line 201)May include surrounding context.

js
});

test('argv: a hostile notebook title is one argument, not shell syntax', () => {
  const argv = cli.argvNoteGet('x"; rm -rf / #', 'p');
  assert.deepEqual(argv, ['-p', 'p', 'note', 'get', 'x"; rm -rf / #']);
  assert.equal(argv.length, 5, 'the injection attempt stays a single argv element');
});

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
95% 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 · test/cron.test.js (reported line 68)May include surrounding context.

js
});

test('R13: a hostile schedule string stays one argv element', () => {
  const argv = cron.buildCronAdd({ cron: '0 7 * * * ; rm -rf /' });
  assert.equal(argv.filter((a) => a.includes('rm -rf')).length, 1);
});

Chaining Abuse

High
Category
Tool Misuse
Confidence
75% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · test/cron.test.js (reported line 68)May include surrounding context.

js
});

test('R13: a hostile schedule string stays one argv element', () => {
  const argv = cron.buildCronAdd({ cron: '0 7 * * * ; rm -rf /' });
  assert.equal(argv.filter((a) => a.includes('rm -rf')).length, 1);
});

Chaining Abuse

High
Category
Tool Misuse
Confidence
75% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · test/nlm-cli.test.js (reported line 201)May include surrounding context.

js
});

test('argv: a hostile notebook title is one argument, not shell syntax', () => {
  const argv = cli.argvNoteGet('x"; rm -rf / #', 'p');
  assert.deepEqual(argv, ['-p', 'p', 'note', 'get', 'x"; rm -rf / #']);
  assert.equal(argv.length, 5, 'the injection attempt stays a single argv element');
});

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · test/cron.test.js (reported line 116)May include surrounding context.

js
});

// ---------------------------------------------------------------------------
// list and remove
// ---------------------------------------------------------------------------

test('R13: list asks for json and all jobs; remove is positional by id', () => {

YARA rule 'agent_skill_destructive_autonomous_actions': Autonomous destructive filesystem, shell history, or repository actions in AI agent skills [agent_skills]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · test/nlm-cli.test.js (reported line 201)May include surrounding context.

js
test: true }), ['-p', 'spike', 'auth', 'check', '--test']);
  assert.deepEqual(cli.argvDoctor('spike'), ['-p', 'spike', 'doctor']);
  assert.deepEqual(
    cli.argvDownloadReport('r-1', '/tmp/r.md', 'spike'),
    ['-p', 'spike', 'download', 'report', 'r-1', '/tmp/r.md'],
  );
});

test('argv: a hostile notebook title is one argument, not shell syntax', () => {
  const argv = cli.argvNoteGet('x"; rm -rf / #', 'p');
  assert.deepEqual(argv, ['-p', 'p', 'note', 'get', 'x"; rm -rf / #']);
  assert.equal(argv.length, 5, 'the injection attempt stays a single argv element');
});

test('argv: a missing id throws rather than building a command that means something else', () => {
  assert.throws(() => cli.argvNoteGet('', 'p'), /noteId is required/);
  assert.throws(() => cli.argvNoteList(null, 'p'), /notebookId is required/);
  assert.throws(() => cli.argvDownloadReport('r', '', 'p'), /destination is required/);
});

test('argv: no profile omits the flag entirely rather than passing an empty on

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
100% 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 · test/nlm-cli.test.js (reported line 202)May include surrounding context.

js
test('argv: a hostile notebook title is one argument, not shell syntax', () => {
  const argv = cli.argvNoteGet('x"; rm -rf / #', 'p');
  assert.deepEqual(argv, ['-p', 'p', 'note', 'get', 'x"; rm -rf / #']);
  assert.equal(argv.length, 5, 'the injection attempt stays a single argv element');
});

Chaining Abuse

High
Category
Tool Misuse
Confidence
75% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · test/nlm-cli.test.js (reported line 202)May include surrounding context.

js
test('argv: a hostile notebook title is one argument, not shell syntax', () => {
  const argv = cli.argvNoteGet('x"; rm -rf / #', 'p');
  assert.deepEqual(argv, ['-p', 'p', 'note', 'get', 'x"; rm -rf / #']);
  assert.equal(argv.length, 5, 'the injection attempt stays a single argv element');
});

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · test/skill-md.test.js (reported line 336)May include surrounding context.

js
assert.match(trigger, /remove <id>/, 'removal is positional by id');
});

test('the cli contract documents the jar location and the never-print rule', () => {
  const doc = DOCS['references/cli-contract.md'];
  assert.match(doc, /\$NOTEBOOKLM_HOME\/profiles\/<profile>\/storage_state\.json/);
  assert.match(doc, /Path, presence, size, mtime and a sha256 digest/);