Back to skill

Security audit

Signal Skills: podcast

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report undeclared capabilities, privacy and network risks, and incomplete inspection of one referenced file; they do not establish malicious intent.

Consider whether you trust the skill to access analytics data, handle local records, make network requests, and publish reports. Its reported publishing and data-access instructions exceed the stated purpose, and one referenced file was only partially inspected, so the evidence is incomplete.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
resolve.py
Finding

Proxy mode permits attacker-controlled transcript URLs to reach internal services

Content
View full analysis

Impact

An attacker can induce HTTP GET requests to internal services accessible to the proxy and have successful text responses saved as transcripts. This can expose internal content to subsequent extraction agents or trigger GET-based actions. No attacker-directed response exfiltration is established.

other

Warning
Location
SKILL.md
Finding

Standing instructions direct publication without a user request

Content
View full analysis

Impact

Unrequested publication can disclose the user's tracked shows, selected predictions, and grading history. scoreboard_report.py also includes absolute local state and script paths in report footers (lines 110-111). The source material being public does not authorize disclosure of the user's research corpus.

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (24)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The supplied commands expose subscription-management operations rather than transcript retrieval or analysis. Managing ongoing subscriptions is a distinct user-facing capability, not merely an implementation detail of on-demand transcript extraction. The ledger.py implementation is not supplied, so storage behavior, network access, and other side effects cannot be established.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The supplied script's primary function is prediction-performance reporting, a capability not disclosed in the description. It invokes ledger.py to obtain scoreboard data and recorded predictions, assigns hit-rate tiers, includes confirmed and failed examples, and renders reports to disk. Although related to previously mined podcast material, this is a distinct analytical output rather than merely an implementation detail of transcript retrieval or idea extraction. The visible code does not fetch transcripts, accept podcast sources, make LLM calls, or publish reports over the network.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
87% confidence
Finding

The supplied commands expose a scoreboard-reporting workflow rather than podcast transcript retrieval or content analysis. Nothing shown connects the scoreboard to the declared podcast workflow. The script implementation is not supplied, so its internal data sources and any additional behavior cannot be established.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
86% confidence
Finding

The supplied command concerns tracking whether a podcast prediction was confirmed or failed using a cited source. Prediction verification and ledger management are not disclosed in the transcript-retrieval and insight-mining description and extend beyond ordinary implementation details of those tasks. The script implementation is absent, so the snippet does not establish whether it fetches the source URL or how it persists results.

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 73)May include surrounding context.

md
Cross-reference this against the episode list `resolve.py` returns (Step 1)

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 83)May include surrounding context.

md
stop — no `resolve.py`, no transcripts, no extractor fan-out:

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 106)May include surrounding context.

md
subscribed source, run `resolve.py` (Step 1), diff against `mined` (this

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 116)May include surrounding context.

md
`resolve.py` (next to this file) does the whole first mile for ANY show.

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 142)May include surrounding context.

md
**YouTube gets its own path in `resolve.py`** — it isn't RSS, so it

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 151)May include surrounding context.

md
`resolve.py` — there is no separate Step 2 fallback needed for YouTube,

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 160)May include surrounding context.

md
Fallback order when `resolve.py` prints no `podcast:transcript` tags:

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 494)May include surrounding context.

md
- Step 1 — `resolve.py` (feed resolution, RSS parsing, `yt-dlp` metadata),

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 520)May include surrounding context.

md
it into `resolve.py` / `ledger.py` rather than doing it inline as a

The skill discovers existing analytics credentials/connections and uses them to query private product data, beyond its stated podcast-analysis purpose.

High
Category
Excessive Agency
Confidence
98% confidence
Finding

The instructions use available analytics connections or credentials as grounds to silently query private product data, without requiring authorization for this podcast-analysis task. Credential availability is not consent, and externally sourced podcast claims can consequently drive access to sensitive business metrics that are then incorporated into reports.

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.

The skill directs unsolicited public republication beyond its stated on-demand podcast-mining purpose.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The skill explicitly encourages publishing even when nobody requested it and republishing after grading passes. This exceeds authorization for on-demand analysis and could expose the user's tracked shows, private annotations, grading decisions, or local paths if an external publishing tool is available.

Content

No source excerpt is available for this finding.

Reframes optional public publishing as standing authorization to republish without a user request.

Medium
Category
Prompt Injection
Confidence
98% confidence
Finding

Although the section is labeled optional, its instructions turn publication into a standing distribution loop without obtaining user consent. Treating skill-authored encouragement as user authorization can cause repeated unwanted disclosures and public attribution of unreviewed rankings.

Content

No source excerpt is available for this finding.

Tainted flow: 'data_dir' from os.environ.get (line 36, credential/environment) → shutil.copy2 (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · ledger.py (reported line 46)May include surrounding context.

python
continue
        data_dir.mkdir(parents=True)
        if has_config:
            shutil.copy2(old / "config.json", data_dir / "config.json")
        if has_state:
            shutil.copytree(old / "state", data_dir / "state")
        return old

Tainted flow: 'path' from os.environ.get (line 101, credential/environment) → open (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · ledger.py (reported line 67)May include surrounding context.

python
def save_json(path, data):
    path.parent.mkdir(parents=True, exist_ok=True)
    with open(path, "w") as f:
        json.dump(data, f, indent=2, sort_keys=True)
        f.write("\n")

Tainted flow: 'output' from sys.stdin.read (line 426, user input) → pathlib.Path.write_text (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · render_report.py (reported line 431)May include surrounding context.

python
print(f"payload error: {e}", file=sys.stderr)
            sys.exit(1)

        dst.write_text(output)
        print(f"wrote {dst} ({len(output)} bytes, {len(payload['tiers'])} tiers)")

        if should_open and not opened and dst.suffix.lower() not in TEXT_EXTENSIONS:

YouTube fetches bypass the documented direct-connection and public-address guarantees.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

YouTube metadata and caption requests run inside yt-dlp, outside check_url, the guarded socket implementation, and the explicit proxy opt-in policy. Consequently, those requests can inherit proxy settings and lack this script's connect-time private-address checks, potentially reaching internal services through manipulated DNS or downstream request destinations; hostname-based YouTube routing narrows exposure but does not enforce the advertised network boundary.

Content

No source excerpt is available for this finding.

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 · resolve.py (reported line 213)May include surrounding context.

python
def yt_dlp_json(url, extra=()):
    cmd = ["yt-dlp", "--skip-download", "--no-warnings", "-J", *extra, url]
    try:
        raw = subprocess.run(cmd, capture_output=True, check=True, timeout=60).stdout
    except FileNotFoundError:
        sys.exit("yt-dlp not found on PATH. Install with `brew install yt-dlp`.")
    except subprocess.CalledProcessError as ex:

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 · resolve.py (reported line 254)May include surrounding context.

python
cmd = ["yt-dlp", "--skip-download", "--no-warnings", sub_flag,
               "--sub-lang", "en.*", "--sub-format", "vtt",
               "-o", base + ".%(ext)s", url]
        subprocess.run(cmd, capture_output=True, timeout=120)
        for fname in os.listdir(workdir):
            if fname.startswith(os.path.basename(base)) and fname.endswith(".vtt"):
                path = os.path.join(workdir, fname)

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 · scoreboard_report.py (reported line 43)May include surrounding context.

python
def _run_ledger(*args):
    out = subprocess.run(
        [sys.executable, str(LEDGER), *args], capture_output=True, text=True, check=True
    ).stdout
    return json.loads(out)