Back to skill

Security audit

Smart Daily Report by YQG

Security checks for vulnerabilities and agentic risk

Overview

The skill is a legitimate work-report helper, but it may search broad local folders and connected work data without a clear consent gate.

Install only if you are comfortable with the agent using local repository history and optional task or calendar data to draft reports. Before using it, specify the exact repo or folder to analyze, avoid broad home or desktop scans, and confirm any Feishu or file export destination before writing the report.

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:38
Finding
Overbroad Collection of Local Project, Task, Identity, and Configuration Data## Vulnerability Details **File Location**: `SKILL.md`, lines 38–55 and 200–206 **Vulnerability Type**: Excessive local data access and insufficient scope restriction **Risk Level**: Medium ### Vulnerable Code `SKILL.md`, lines 38–55: ```markdown **Project commit history:** - Use appropriate system tool to read commit log for the specified date range - Collect: commit message, timestamp, project name - Include all branches the user contributed to **Change statistics:** - Collect number of files changed, lines added/removed - Group by project if multiple repos **Task completion:** - If todoist CLI is available, query completed tasks for the date - Otherwise, skip this source ### Multi-Project Handling If the user has multiple projects: 1. Scan for project directories in common locations (workspace, home, desktop) 2. Collect activity from each 3. Group results by project in the output ``` `SKILL.md`, lines 200–206: ```markdown | User provides manual input | Combine manual descriptions with any available stats | | Multiple contributors detected | Filter to only the user's commits (by author name/email) | | Very active day (20+ commits) | Group aggressively, focus on features/fixes, collapse minor commits into stats | | Across time zones | Use the user's local date (from USER.md timezone setting) | ``` ### Technical Analysis The Skill instructs the Agent to scan broad filesystem locations, including the user's home and desktop directories, instead of limiting collection to repositories explicitly selected for the report. It also directs the Agent to inspect all contributed branches, query Todoist data, derive identity from author names or email addresses, and read timezone information from `USER.md`. These operations may be performed merely because the user requested a work report. The instructions do not require separate informed consent for each data source, define an allowed repository ro ...[truncated 2409 chars]
Remediation
## Remediation Suggestions 1. Default collection to the current working repository or a repository path explicitly supplied by the user. 2. Require explicit confirmation before searching additional directories, accessing task managers, reading calendars, or consulting user-profile files. 3. Replace automatic scanning of home and desktop directories with a user-provided allowlist of repository roots. 4. Apply strict search boundaries, including maximum directory depth, repository count, and excluded paths. 5. Do not read `USER.md` solely to determine timezone. Ask the user for the relevant timezone or use a non-sensitive runtime setting with their consent. 6. Ask the user to provide the Git author identity to filter, rather than automatically collecting names and email addresses from repository history. 7. Limit Git inspection to the requested date range, selected branches, and selected repositories. 8. Make Todoist and calendar access opt-in and clearly state what fields will be collected before invoking those integrations. 9. Redact email addresses, private repository names, task descriptions, and other sensitive metadata unless the user explicitly requests their inclusion. 10. Present a data-source preview before collection, for example: “Use the current repository only,” “Add selected repositories,” or “Include task and calendar data.”
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (6)

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The description presents the skill as a report generator but does not warn that it may inspect commit history, task-manager data, calendar activity, and potentially discover multiple projects on disk. This weak disclosure undermines informed consent and makes the skill more dangerous because users may not realize the breadth of local and connected data access involved.

Vague Triggers

Medium
Confidence
93% confidence
Finding
Several trigger phrases like 'What did I work on' or short report-related phrases are generic enough to collide with normal chat requests. In isolation this is not harmful, but paired with data-collection behavior it raises the risk of unintended activation and unnecessary access to local or linked data sources.

Vague Triggers

Medium
Confidence
97% confidence
Finding
The trigger phrase 'Any request to summarize work activity' is overly broad and can activate on ordinary conversation that does not imply consent to inspect local repos, task systems, or calendars. Overbroad activation increases the chance that the skill gathers sensitive data in contexts where the user only wanted lightweight assistance.

Context-Inappropriate Capability

Medium
Confidence
91% confidence
Finding
The skill explicitly instructs scanning common local directories such as workspace, home, and desktop to discover projects, which expands data collection beyond user-provided or clearly scoped repositories. That behavior can expose unrelated repositories, confidential source code metadata, and private work context without an explicit, narrow consent step.

Description-Behavior Mismatch

Medium
Confidence
94% confidence
Finding
The export options permit writing generated reports to external services and local files, which changes the skill from summarization to persistence and potential data exfiltration. Because reports may contain commit summaries, task details, and calendar-derived information, exporting without strong consent and destination transparency can leak sensitive operational data.

Missing User Warnings

Low
Confidence
88% confidence
Finding
The skill describes saving reports to files or external document services without warning that this creates persistent copies or may share sensitive work summaries outside the chat context. Even if user-directed, the absence of a persistence/sharing warning can lead to accidental exposure of internal project information.

Static analysis

No suspicious patterns detected.