Back to skill

Security audit

Agent Daily Review

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent daily-review helper, but it can read more memory than advertised and persist summaries by default, so users should review it before installing.

Install only if you are comfortable with the skill reading your workspace memory and writing persistent review files. Prefer running it manually first with --no-memory, review the generated report before sharing it, and avoid scheduled cron use until the MEMORY.md date filtering is tightened and persistence is made explicit.

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/daily_review.py:48
Finding

Daily Review Collects Historical Memory Entries Outside the Requested Date

Content
View full analysis
List[str]: """Extract decision records""" decisions = [] # Find lines containing decision-related keywords decision_keywords = ['decision', 'decided', 'chose', 'confirmed', 'adopted'] for line in content.split('\n'): if date in line or any(kw in line.lower() for kw in decision_keywords): if any(kw in line.lower() for kw in decision_keywords): decisions.append(line.strip()) return decisions[:10] # Limit quantity def _extract_learnings(self, content: str, date: str) -> List[str]: """Extract learning/lesson records""" learnings = [] learning_keywords = ['learned', 'lesson', 'experience', 'note', 'improve'] for line in content.split('\n'): if date in line or any(kw in line.lower() for kw in learning_keywords): if any(kw in line.lower() for kw in learning_keywords): learnings.append(line.strip()) return learnings[:10] ``` ### Technical Analysis The review workflow reads the complete `MEMORY.md` file and passes its contents to extraction functions that are intended to identify entries associated with the requested review date. However, the outer condition in each extraction function is logically ineffective as a date restriction: ```python if date in line or any(keyword_matches): ``` A line is accepted whenever it contains a ...[truncated 2574 chars]
Remediation
View remediation
list[str]: lines = content.splitlines() selected = [] in_target_section = False for line in lines: heading = re.match(r'^##\s+(\d{4}-\d{2}-\d{2})(?:\s+Review)?\s*$', line.strip()) if heading: in_target_section = heading.group(1) == date continue if in_target_section: selected.append(line) return selected ``` Keyword extraction should then operate exclusively on the returned target section. ]]>
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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
Findings (7)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding

The skill clearly instructs reading from memory files and workspace artifacts and writing reports to reviews plus appending to MEMORY.md, but it declares no explicit tool scope or permissions. That mismatch can cause overbroad execution in hosts that rely on manifest-declared capabilities, reducing transparency and increasing the chance of unintended file access or persistence.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger phrase "Summarize today" is broad and likely to overlap with ordinary user conversation, which can cause accidental invocation of a skill that scans local records and writes review output. Because this skill performs both read and write actions, an unintended trigger has privacy and integrity consequences beyond a harmless response mix-up.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The usage and workflow indicate that the skill saves a review report and appends a summary to long-term memory, but the skill description does not prominently warn users about this persistence behavior. Users may reasonably expect a summary-only action, so silent writes can expose sensitive notes, create unwanted records, or contaminate memory with inaccurate generated content.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The cron examples enable unattended execution that scans records and persists reports without interactive confirmation or a warning about the privacy implications. Scheduled autonomous reads and writes increase risk because they can repeatedly process sensitive data, create unwanted archives, and do so when the user is not actively supervising the action.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The script appends to MEMORY.md by default, causing persistent modification of long-term records without requiring explicit user opt-in. In an agent skill context, automatic state mutation is more dangerous because background or scheduled runs can silently pollute memory, duplicate entries, and influence future agent behavior based on unintended or low-quality summaries.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The natural-language/config example fixes a specific timezone, which imposes a locale assumption on all users. The file does not present this as an example to customize or provide any opt-in or justification for the region-specific default.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

This code writes a generated markdown report to the user's workspace, which is a file-modifying operation. Although there is a console message after saving, there is no warning before execution in the function docstring or comments that running the workflow will create or overwrite a report file.

Content

No source excerpt is available for this finding.