Back to skill

Security audit

Giggle Video

Security checks for vulnerabilities and agentic risk

Overview

This video-generation skill is mostly coherent, but it needs Review because it uses a bearer API key with a configurable request origin and dynamically loads platform instructions without clear allowlisting.

Install only if you are comfortable sending prompts and video parameters to giggle.pro and storing generated-output metadata locally. Use a narrowly scoped Giggle API key, avoid setting GIGGLE_API_BASE unless you fully trust and validate the destination, and prefer explicit platform selection until the skill adds a fixed platform allowlist.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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)

T01 · Skill Instruction Hijacking

Warning
Location
SKILL.md:28
Finding

User-Controlled Platform Identifier Can Load Unintended Skill Instructions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 28-48
Vulnerability Type: Path traversal into instruction files
Risk Level: Medium

Vulnerable Code

markdown
| prompt       | string | Yes | Video description/prompt                         |
| platform     | string | No  | Target platform ID (see impl/), default `giggle` |
markdown
Stage 2 — Submit task
  Load impl/{platform}.md according to the platform parameter.
  Invoke the implementation layer to submit a generation task and obtain task_id.

Technical Analysis

The platform parameter is supplied by the user and is interpolated directly into the instruction-file path impl/{platform}.md. The skill does not require the Agent or runtime to validate the value against a fixed platform allowlist, canonicalize the resulting path, or reject path separators and traversal sequences.

A platform value containing ../, an absolute path, or platform-specific separators could therefore resolve outside the intended impl/ directory. Because the loaded Markdown is treated as implementation instructions, loading an unintended file can alter the Agent's behavior within the current session. This is best classified as instruction hijacking rather than ordinary data-file disclosure.

Attack Path

  1. An attacker asks the Agent to generate a video and supplies a crafted platform value, such as ../SKILL or another traversal path targeting a Markdown file.
  2. The Agent follows the documented construction rule and attempts to load impl/{platform}.md.
  3. Path normalization resolves the crafted path outside the intended implementation directory.
  4. The Agent reads the targeted file as trusted implementation guidance.
  5. Instructions in that file can redirect the current workflow, request additional tools, or override the intended video-generation procedure.

Exploitation requires a suitable readable file to exist at a path reachable ...[truncated 643 chars]

Remediation
View remediation

Remediation Suggestions

  • Replace path interpolation with a fixed mapping, for example giggle -> impl/giggle.md.
  • Reject unknown platform identifiers before opening any file.
  • Restrict identifiers to a conservative pattern such as ^[a-z0-9_-]+$.
  • Explicitly reject /, \, .., URL schemes, absolute paths, and encoded path separators.
  • Resolve and canonicalize the selected path, then verify that it remains under the canonical impl/ directory.
  • Treat implementation documents as package-controlled resources rather than arbitrary user-selected files.

T09 · Insecure Skill Coding Practices

Error
Location
impl/giggle.md:9
Finding

Configurable API Base URL Can Exfiltrate the Bearer Credential

Content
View full analysis

Vulnerability Details

File Location: impl/giggle.md, lines 9-21
Vulnerability Type: Unvalidated credential-bearing request destination
Risk Level: High

Vulnerable Code

markdown
| platform id       | `giggle`                                              |
| Network allowlist | `api.giggle.pro`, `cdn.giggle.pro`                    |
| Authentication    | Bearer Token                                          |
| API Key variable  | `GIGGLE_API_KEY`                                      |
| Base URL variable | `GIGGLE_API_BASE` (default `https://api.giggle.pro/v1`) |

## Submit generation task

POST ${GIGGLE_API_BASE}/video/generate Headers: Authorization: Bearer ${GIGGLE_API_KEY} Content-Type: application/json

text

Technical Analysis

The destination of the API request is controlled by the GIGGLE_API_BASE environment variable, while the same request unconditionally carries GIGGLE_API_KEY in its Authorization header. Although a safe default and a domain allowlist are documented, no validation procedure requires the resolved request URL to use HTTPS and the exact hostname api.giggle.pro.

Consequently, a modified environment value can redirect the request to an attacker-controlled server. The request would disclose both the bearer credential and user-provided generation content. The configurable destination also conflicts with the stated constraint in SKILL.md that only allowlisted implementation domains may be accessed.

Redirect handling presents a related risk if the HTTP client forwards authorization headers across origins. DNS and URL parsing ambiguities should also be considered when implementing the hostname check.

Attack Path

  1. An attacker or compromised deployment mechanism changes GIGGLE_API_BASE to an attacker-controlled endpoint, such as https://attacker.example/v1.
  2. A user invokes the video-generation skill.
  3. The Agent constructs the gene ...[truncated 967 chars]
Remediation
View remediation

Remediation Suggestions

  • Prefer a hardcoded API origin and allow configuration only for non-sensitive path components.
  • If an override is operationally necessary, parse it as a URL and require:
    • the https scheme;
    • the exact hostname api.giggle.pro;
    • an approved port;
    • no embedded user information;
    • an approved base path.
  • Do not validate destinations with substring or suffix-only comparisons.
  • Disable cross-origin forwarding of the Authorization header and reject redirects to non-allowlisted origins.
  • Apply the same validation to generation, estimation, and status-polling requests.
  • Use a narrowly scoped API key, rotate it after suspected exposure, and monitor audit logs for unauthorized use.
  • Enforce the documented network allowlist at the gateway or network-egress layer rather than relying only on skill instructions.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (6)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill description does not clearly disclose that prompts and generation metadata will be transmitted to third-party video platforms. Users may include sensitive business, personal, or copyrighted material in prompts without understanding that the data leaves the local environment, creating privacy and confidentiality risks.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger conditions are broad enough to match generic requests like generating videos or discussing creative ideas, which can cause the skill to activate outside a narrowly intended context. Over-broad activation increases the chance that user content is sent to an external service or archived without the user clearly intending to use this specific integration.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documented archive path shows that generated results and metadata are stored in a shared/local output location, but the skill does not warn users of this persistence behavior. This can expose prompts, URLs, and project metadata to other local users, processes, or downstream tooling that has access to the shared directory.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · impl/giggle.md (reported line 13)May include surrounding context.

md
| 网络白名单        | `api.giggle.pro`, `cdn.giggle.pro`                    |
| 认证方式          | Bearer Token                                          |
| API Key 环境变量  | `GIGGLE_API_KEY`                                      |
| Base URL 环境变量 | `GIGGLE_API_BASE`(默认 `https://api.giggle.pro/v1`) |

## 提交生成任务

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

Most of the operational instructions, triggers, parameters, and safety constraints are presented only in Chinese, which effectively imposes a language requirement on users and maintainers. The file does not indicate that this locale is optional, user-selectable, or justified as a region-specific skill.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

This markdown file documents authentication via a Bearer token and names the GIGGLE_API_KEY environment variable, but it does not include any warning about handling credentials securely or avoiding accidental exposure. Under the markdown-file criteria for missing user warnings, this is a privacy/safety-relevant behavior that should be disclosed to users.

Content

No source excerpt is available for this finding.