Back to skill

Security audit

Daily Shortdrama Report

Security checks for vulnerabilities and agentic risk

Overview

This skill generates a Chinese daily short-drama work report and its local memory reads are disclosed and purpose-aligned, though users should be aware it may read recent memory files when invoked.

Install this if you want a Chinese daily short-drama production report. Before using it, make sure memory/YYYY-MM-DD.md and memory/targets.md contain only information suitable for a report, or ask the agent to confirm and redact sensitive details before generating output.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:13
Finding

Overbroad Access to Persistent Agent Memory

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 13-21
Vulnerability Type: T05: Unauthorized Access and Privilege Escalation
Risk Level: Medium

Vulnerable Code

markdown
### Step 2: Read today's content-production records
Read the current and previous day's content-production data from `memory/YYYY-MM-DD.md`:
- Number of completed scripts, storyboards, and generated videos
- Platform publishing activity
- Team collaboration events

### Step 3: Read targets for comparison
Read `memory/targets.md`, if it exists, to obtain weekly or monthly targets and compare them with actual completion.

The snippet above is an English rendering of the corresponding source instructions.

Technical Analysis

The Skill instructs the Agent to read complete persistent-memory files for the current day, previous day, and team targets. Although the requested fields are relevant to producing a daily report, the instructions do not restrict access to structured fields, require user authorization, validate the selected memory paths, or require redaction of unrelated content.

Daily memory files may contain private conversation context, personal information, credentials, internal business details, or other records unrelated to production statistics. Reading the whole files places all such content in the Agent's active context and creates a risk that it may be incorporated into the generated report. This violates least-privilege principles because the Skill needs specific production metrics rather than unrestricted access to all content in those files.

No evidence indicates that the Skill modifies memory, obtains operating-system privileges, transmits information externally, or accesses arbitrary paths. The issue is therefore limited to overbroad access and potential disclosure within the generated report.

Attack Path

  1. Sensitive or unrelated information is stored in the current day's, previous day's, or t ...[truncated 1089 chars]
Remediation
View remediation

Remediation Suggestions

  1. Require explicit user authorization before reading persistent-memory files.
  2. Replace free-form memory access with a structured data source containing only report fields such as completed scripts, storyboards, generated videos, publishing activity, and collaboration events.
  3. Enforce an exact path allowlist and reject path traversal, substitutions, or references outside the designated reporting files.
  4. Parse only explicitly approved fields instead of loading complete file contents into the reporting context.
  5. Add mandatory redaction for credentials, tokens, personal data, private conversation content, and unrelated business information.
  6. Instruct the Agent never to reproduce memory content verbatim unless the user explicitly requests and authorizes it.
  7. If a required field cannot be retrieved safely, preserve the existing behavior of displaying an unavailable-value marker rather than broadening access.
  8. Limit report distribution to authorized recipients and review generated reports for sensitive content before publication.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger phrases in the metadata include very common terms such as “日报”, “每日总结”, and “每日报告”, which are likely to appear in ordinary conversation. This can cause unintended skill activation, leading the agent to read session statistics and local memory files without sufficiently explicit user intent, increasing the risk of unnecessary data exposure or confusing behavior.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The dedicated trigger list is similarly broad and ambiguous, with phrases that map to generic reporting requests rather than this specific skill. In context, this skill is allowed to pull usage and productivity data from session_status and memory files, so accidental activation has a meaningful privacy and scope impact beyond merely formatting text.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The skill description, report template, and trigger phrases are all fixed in Chinese, and there is no indication that the user may choose another language or locale. Under the stated policy, forcing a specific language without opt-in can be a natural-language policy issue.

Content

No source excerpt is available for this finding.