Back to skill

Security audit

Agent Escalation (Webhook)

Security checks for vulnerabilities and agentic risk

Overview

The skill is a disclosed escalation helper, but it forwards potentially sensitive troubleshooting context and a bearer token through an under-scoped webhook path to a fixed Telegram destination.

Install only if this skill is meant for your organization’s escalation workflow. Configure and verify the gateway, token, supervisor, channel, and recipient before use; keep the gateway on trusted loopback or require TLS for any non-local destination; and do not include passwords, API keys, tokens, personal data, or raw customer logs in escalation arguments.

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 (3)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/escalate.sh:10
Finding

Bearer Token and Escalation Data Exposed Through Configurable Plaintext HTTP Destination

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/escalate.sh:13
Finding

JSON Injection Through Unescaped Environment-Controlled Routing Fields

Content
View full analysis
Remediation
View remediation

other

Warning
Location
scripts/escalate.sh:13
Finding

Sensitive Escalation Context Is Automatically Delivered to a Fixed Telegram Recipient

Content
View full analysis
/scripts/escalate.sh "" "" "" ``` ### Technical Analysis The script sets `deliver` to `true`, defaults the delivery channel to Telegram, and embeds a fixed numeric recipient identifier. It forwards the complete problem description, attempted remediation, and result without redaction or a sensitivity check. Troubleshooting data commonly contains access tokens, customer information, internal service URLs, account identifiers, stack traces, and infrastructure details. The implementation provides no confirmation step, recipient verification, data minimization, or secret filtering before requesting external delivery. The project documentation indicates that Telegram delivery is intended, so this is not hidden functionality. The security weakness is the absence of safeguards around potentially sensitive content and the use of a package-level fixed recipient. ### Attack Path 1. A user or agent encounters a technical problem involving sensitive customer or infrastructure information. 2. The escalation arguments include an error message, token, internal endpoint, customer identifier, or other confidential context. 3. The script combines the supplied values into the escalation message. 4. The webhook reques ...[truncated 908 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (5)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
85% confidence
Finding

The skill instructs the agent to execute a shell script but does not declare any explicit tool scope or permissions. This increases the chance of unintended shell access or overbroad execution in environments that rely on metadata-based permission gating, making the skill harder to audit and safely sandbox.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill says the supervisor receives the complete context and that the response is delivered over Telegram, but it gives no warning or filtering guidance for secrets, tokens, personal data, or client-sensitive material. In an escalation workflow, operators are likely to paste raw errors, credentials, email contents, or customer data, creating a real risk of data leakage to external systems and chat channels.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The script’s natural-language comments, usage strings, and user-facing error text are exclusively in French, such as the usage/help and missing-token message. This imposes a specific language/locale on users without offering a choice or documenting a justified region-specific constraint, matching the language-policy violation criteria.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The script transmits user-supplied problem details and attempt history to an external webhook endpoint, which can expose potentially sensitive operational data. The risk is increased because the request uses plain HTTP rather than HTTPS, so the bearer token and message contents could be intercepted or modified if the connection is not strictly local and trusted.

Content

Scanner excerpt · scripts/escalate.sh (reported line 34)May include surrounding context.

sh
JSON_MESSAGE=$(printf '%s' "$MESSAGE" | python3 -c 'import json,sys; print(json.dumps(sys.stdin.read()))')

# Appel webhook
RESPONSE=$(curl -s -w "\n%{http_code}" -X POST \
  "http://${GATEWAY_HOST}:${GATEWAY_PORT}/hooks/agent" \
  -H "Authorization: Bearer ${HOOKS_TOKEN}" \
  -H "Content-Type: application/json" \

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest description and operational instructions are entirely in French, which effectively forces a specific language for users of the skill. The file does not indicate that French is optional, user-selected, or required for a documented region-specific reason.

Content

No source excerpt is available for this finding.