Back to skill

Security audit

FCPX Assistant

Security checks for vulnerabilities and agentic risk

Overview

This FCPX automation skill has a coherent video-production purpose, but its Web UI and scripts create serious risks including unauthenticated command execution and deletion of unrelated media files.

Treat this as a Review item and do not run the Web UI on a networked machine until it is fixed. Before installing, require removal of shell=True and eval command construction, bind the Web UI to localhost with authentication, disable broad cleanup by default, pin dependencies, avoid forced HTTP downloads, and isolate any publishing cookies or tokens from the automation environment.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (5)

T09 · Insecure Skill Coding Practices

Error
Location
webui/app.py:28
Finding

Unauthenticated Remote Command Execution Through the Web UI

Content
View full analysis
/tmp/webui-rce-proof; # ``` After interpolation, the shell receives a command equivalent to: ```bash bash ".../scene-detect.sh" ""; id > /tmp/webui-rce-proof; #" ``` The injected command is consequently executed by the shell. No input escaping, allowlist validation, canonical path restriction, or argument-arra ...[truncated 1604 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/auto-publish.sh:239
Finding

Command Injection Through Dynamic Bilibili Publishing Command

Content
View full analysis
/tmp/publish-rce-proof; # ``` causes the assembled command to contain executable shell syntax. `eval` reparses the entire value and executes the injected command. This vulnerability can be reached directly through the command-line script and indirectly through the unauthenticated Web UI publishing form. ### Attack Path 1. The attacker supplies a crafted publishing title, description, tag, cover path, schedule, or video path. 2. `webui/app.py` or a direct script invocation forwards the value to `auto-publish.sh`. 3. The script stores the value in a shell variable. 4. `upload_bilibili()` concatenates it into `BILI_CMD`. 5. `eval "$BILI_CMD"` reparses the attacker-controlled string as shell code. 6. The injected command executes with the privileges of th ...[truncated 674 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/auto-video-maker.sh:496
Finding

Default Cleanup Deletes Unrelated Videos and Subtitle Files

Content
View full analysis
/dev/null && \ echo -e " Deleted trial video files" # Delete old subtitle files find "$(dirname "$OUTPUT_FILE")" -maxdepth 1 -name "*.srt" -type f -delete 2>/dev/null && \ echo -e " Deleted old subtitle files" # Delete temporary voiceover if [[ -d "$PROJECT_DIR/voiceover" ]] && [[ "$PROJECT_DIR/voiceover" != "$VOICEOVER_DIR" ]]; then trash "$PROJECT_DIR/voiceover" 2>/dev/null || rm -rf "$PROJECT_DIR/voiceover" echo -e " Deleted old voiceover files" fi fi ``` ### Technical Analysis The script does not track which files were created during the current run. Instead, it treats every sibling MP4 other than the selected output and every sibling SRT as disposable. If `OUTPUT_FILE` is located in a shared media directory, all other top-level MP4 and SRT files in that directory are deleted. These files may be unrelated source videos, completed exports, or manually maintained subtitles. The voiceover cleanup also recursively removes `$PROJECT_DIR/voiceover` when it differs textually from `$VOICEOVER_DIR`. The comparison is not based on canonical paths and does not establish that the directory was generated by the current execution. Because automatic cleanup defaults to enabled, users need to know about and explicitly select `--no-auto-cleanup` to avoid this behavior. ### Attack Path 1. A user or ...[truncated 1012 chars]
Remediation
View remediation
/dev/null && pwd -P)" selected_voiceover="$(cd "$VOICEOVER_DIR" 2>/dev/null && pwd -P)" ``` 6. Refuse cleanup if a target is empty, `/`, the user's home directory, or outside the dedicated project root. 7. Prompt for explicit confirmation before deleting non-temporary assets. 8. Prefer recoverable deletion, and do not silently fall back to `rm -rf`. 9. Add integration tests that place unrelated MP4 and SRT files beside the output and verify that they remain intact. ]]>

T08 · Insecure Dependencies

Warning
Location
start-webui.sh:72
Finding

Unpinned Dependencies Are Automatically Upgraded on Every Launch

Content
View full analysis
"$REQUIREMENTS_FILE" << EOF gradio>=4.0.0 numpy>=1.20.0 pillow>=9.0.0 EOF # Install dependencies pip3 install -q -r "$REQUIREMENTS_FILE" --upgrade ``` ### Technical Analysis The launcher rewrites the project's dependency manifest and requests the newest version satisfying each lower bound on every startup. No exact version pins, integrity hashes, or audited lock file are used. The resulting environment depends on the state of the package index at launch time. A future malicious or compromised eligible release, dependency takeover, or unexpectedly vulnerable transitive release can be installed without a source-code change to this project. Python packages and their build backends can execute code during installation. Therefore, dependency installation is itself a code-execution boundary and should not occur implicitly during routine application startup. ### Attack Path 1. A compromised, malicious, or vulnerable release satisfying one of the broad version constraints becomes available from the configured package index. 2. A user executes `start-webui.sh`. 3. The launcher overwrites `requirements.txt` with lower-bound-only constraints. 4. `pip3 install --upgrade` resolves and downloads the new package or transitive dependency. 5. Installation-time code runs, or the unsafe package is imported when the Web UI starts. 6. The package receives the privileges and filesystem access of the user launching the application. ### Impact Assessment The direct impact depends on the compromised dependency. A malicious package may obtain arbitrary code execution as the current user, including access to: - Project and media files. - Environment variables containing API credentials. - Publish ...[truncated 289 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
scripts/music-collector.sh:118
Finding

Music Downloads Are Forcibly Downgraded to Unauthenticated HTTP

Content
View full analysis
/dev/null || echo "[]") HITS=$(echo "$RESP" | jq 'length' 2>/dev/null || echo "0") if [[ "$HITS" -eq 0 || "$RESP" == "[]" ]]; then SEARCH_Q=$(echo "$KEYWORDS" | sed 's/ /+/g') RESP=$(curl -s "http://ccmixter.org/api/query?search=${SEARCH_Q}&limit=${COUNT}&f=json" --connect-timeout 10 --max-time 30 2>/dev/null || echo "[]") HITS=$(echo "$RESP" | jq 'length' 2>/dev/null || echo "0") fi ``` HTTPS download URLs are deliberately changed to HTTP and then followed: ```bash # Obtain MP3 download URL DL_URL=$(echo "$RESP" | jq -r ".[$i].files[] | select(.file_format_info.\"format-name\" | test(\"mp3\")) | .download_url" 2>/dev/null | head -1) # ccMixter SSL certificate has issues, force HTTP DL_URL=$(echo "$DL_URL" | sed 's|^https://|http://|') if [[ -z "$DL_URL" || "$DL_URL" == "null" ]]; then FILE_PAGE=$(echo "$RESP" | jq -r ".[$i].file_page_url // empty" 2>/dev/null) [[ -z "$FILE_PAGE" ]] && continue DL_URL="$FILE_PAGE" fi if curl -sL "$DL_URL" -o "$OUTPUT_FILE" --connect-timeout 15 --max-time 120 2>/dev/null; then ``` ### Technical Analysis HTTP does not authenticate the server or protect response integrity. Any attacker able to intercept or alter traffic can modify both the API response and downloaded file. Because the API response supplies `download_url`, a network attacker can redirect the client to an attacker-controlled host. The `-L` option follows redirects, and the code does not restrict the destination host or protocol after redirection. Downloaded content is accepted primarily based on its size and is subsequently passed to ...[truncated 1418 chars]
Remediation
View remediation
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (99)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

This finding mentions external music sourcing, yt-dlp downloads, and browser launching that are not reflected in the declared FCPX-focused purpose. Undeclared external-network and browser actions can expose users to unexpected data transfer, untrusted content, and policy or licensing issues, especially when embedded in an automation skill.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest contains very broad trigger phrases such as generic video-making, importing, exporting, publishing, and common multilingual terms. Overbroad activation increases the chance the skill is invoked for unrelated user requests, which is especially dangerous here because the skill can lead to shell execution, file operations, and potentially external publishing workflows.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 42)May include surrounding context.

md
bash scripts/music-collector.sh --keywords "轻松 愉快" --count 3 --output ./my-project/music

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 48)May include surrounding context.

md
bash scripts/tts-voiceover.sh --script-file ./script.txt --output ./my-project/voiceover --merge