Back to skill

Security audit

Meeting Coordinator - In Person + Virtual (Google Meet)

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent meeting-scheduling skill with disclosed Google calendar/email access and strong approval gates, though users should be careful with credential scope and verify availability results.

Install only with a dedicated agent Google account and the minimum calendar permissions needed. Before approving calendar changes, verify the active account, calendar ID, recipients, times, and locations, and treat availability results cautiously if the calendar CLI reports errors or unusual output.

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 (1)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/check-availability.py:25
Finding
Fail-Open Calendar Retrieval Can Produce False Availability## Vulnerability Details **File Location**: `scripts/check-availability.py`, lines 25–47 and 66–73 **Vulnerability Type**: Fail-open error handling and incomplete calendar parsing **Risk Level**: Medium ### Vulnerable Code ```python try: result = subprocess.run( ["gog", "calendar", "events", calendar_id, "--from", f"{date_str}T00:00:00", "--to", f"{date_str}T23:59:59"], capture_output=True, text=True, check=True, ) except subprocess.CalledProcessError as e: print(f"Error fetching calendar: {e}", file=sys.stderr) return [] events = [] for line in result.stdout.strip().split("\n"): if not line.strip() or line.startswith("ID"): continue parts = line.split() if len(parts) < 4: continue events.append({ "id": parts[0], "start": parts[1], "end": parts[2], "summary": " ".join(parts[3:]), }) ``` ```python busy = [] for ev in raw_events: try: s = parse_dt(ev["start"], tz) e = parse_dt(ev["end"], tz) busy.append((s, e, ev["summary"], ev["id"])) except Exception as exc: print(f"Warning: skipping event {ev}: {exc}", file=sys.stderr) ``` ### Technical Analysis Calendar retrieval failures are converted into an empty event list. The same effective result occurs when output rows do not match the assumed whitespace-delimited format or when an event timestamp cannot be parsed: affected events are silently omitted. `find_free_slots()` cannot distinguish a genuinely empty calendar from failed or incomplete retrieval. It consequently calculates availability from an empty or partial set of busy periods and may report occupied periods as free. This violates the safer fail-closed behavior expected for scheduling decisions. Although the subprocess invocation uses an argument array and is not vulnerable to shell injection, its output hand ...[truncated 1391 chars]
Remediation
## Remediation Suggestions 1. Fail closed when `gog` exits unsuccessfully. Raise an exception or return an explicit error object instead of returning an empty event list. 2. Prevent free-slot calculation and all downstream scheduling actions unless calendar retrieval completed successfully. 3. Use a structured `gog` output mode such as JSON, if supported, rather than parsing human-readable whitespace-delimited output. 4. Validate every event record, including required fields, timestamp syntax, timezone information, and the condition that the end time is later than the start time. 5. Treat any unparseable event within the requested range as an incomplete availability result rather than silently skipping it. 6. Return explicit status fields such as `availability_status`, `retrieval_complete`, and `errors`, and require callers to verify them. 7. Add tests covering authentication failures, permission failures, malformed rows, all-day events, summaries containing whitespace, timezone offsets, daylight-saving transitions, and partial output.
Vulnerability Patterns
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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 (5)

Natural-Language Policy Violations

Medium
Confidence
94% confidence
Finding
The instructions require displaying times with US-specific labels like ET, CT, MT, and PT, and reinforce that format in examples. This imposes a locale-specific communication standard without user opt-in or a documented justification, which fits the language/locale policy violation category.

Lp3

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding
The skill instructs the agent to use shell-accessible tools (`gog`, `goplaces`, `python3`, local scripts) but does not declare an explicit tool permission boundary such as `permissions` or `allowed-tools`. In a credentialed scheduling context, that increases the risk of overbroad command execution, misuse of local OAuth state, and accidental access to email/calendar data beyond what the user expects.

Natural-Language Policy Violations

Medium
Confidence
96% confidence
Finding
The instructions require email time display to use standard US labels like ET, CT, MT, and PT, including dual-time formatting, as a fixed policy. This is a natural-language locale constraint that forces a specific regional convention rather than offering user choice or documenting a justified region-specific scope.

subprocess module call

Medium
Category
Dangerous Code Execution
Content
def get_events(calendar_id: str, date_str: str, tz: ZoneInfo):
    """Fetch events for a date from gog calendar."""
    try:
        result = subprocess.run(
            ["gog", "calendar", "events", calendar_id,
             "--from", f"{date_str}T00:00:00",
             "--to", f"{date_str}T23:59:59"],
Confidence
70% confidence
Finding
subprocess module calls execute external commands. Without careful input validation, this enables command injection.

subprocess module call

Medium
Category
Dangerous Code Execution
Content
search_term = f"{query} near {location}"

    try:
        result = subprocess.run(
            ["goplaces", "search", search_term],
            capture_output=True, text=True, check=True,
        )
Confidence
70% confidence
Finding
subprocess module calls execute external commands. Without careful input validation, this enables command injection.