Back to skill

Security audit

Ada Agent Autonomy Kit

Security checks for vulnerabilities and agentic risk

Overview

This skill openly teaches agents to keep working through queues and scheduled jobs, but it gives them too much persistent autonomy without clear limits or approval controls.

Install only if you intentionally want autonomous scheduled agent work. Before enabling the cron or heartbeat examples, restrict the session tools and workspace, keep the task queue writable only by trusted users or agents, review tasks before they become Ready, set time and token limits, and know how to list and remove the cron jobs.

Vulnerability Patterns
  • 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
  • System PersistenceInstalls backdoors, hooks, services, or scheduled tasks that survive the run
  • 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)

T06 · System Persistence

Error
Location
README.md:163
Finding

Recurring heartbeat and cron jobs establish persistent autonomous execution

Content
View full analysis
Remediation
View remediation

T02 · Agent Memory Poisoning

Error
Location
templates/HEARTBEAT.md:10
Finding

Writable persistent task state can steer future agent sessions

Content
View full analysis
Remediation
View remediation

T01 · Skill Instruction Hijacking

Warning
Location
SKILL.md:12
Finding

Open-ended continuous-work instructions weaken task authorization boundaries

Content
View full analysis
Remediation
View remediation
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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (9)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The README explicitly promotes unattended autonomous execution ('These run automatically — no human prompt needed') and earlier instructs agents to continuously pick tasks, do work, and update shared resources. In the context of an agent skill, this can lead to unsupervised modifications to repositories, memory files, task queues, external channels, or other connected systems without clear operator consent, scope limits, or safety guardrails.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The description 'Stop waiting for prompts. Keep working.' encourages autonomous behavior without stating boundaries, approval requirements, or explicit activation conditions. In an agent skill, this can cause the agent to continue acting outside user intent or expected session scope, increasing the chance of unauthorized changes, excessive tool use, or unintended persistence.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The phrase 'Transform your agent from reactive to proactive' is marketing language that implies expanded autonomous behavior but does not specify when such behavior is permitted. In context with the rest of the file, especially references to heartbeat-driven work and continuous operation, it normalizes self-directed action that may bypass explicit user prompting or narrowly scoped task execution.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The heartbeat switches into autonomous 'work mode' whenever nothing is marked urgent, but it does not define clear authorization, scope, or stopping conditions for self-initiated work. In an agent skill explicitly designed to 'keep working,' this can cause the agent to take unprompted actions, consume resources, or act on stale task state without explicit user consent.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instruction 'Idle time = wasted tokens. Keep working.' strongly encourages continuous self-invocation and persistence even in the absence of fresh user input. In this skill context, that increases the chance of runaway autonomy, excessive token consumption, repeated actions, and deviation from user intent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The examples set '--tz "America/Vancouver"' and repeat that locale in multiple cron commands, which imposes a specific regional setting in natural-language usage guidance. Because the README does not present this as a placeholder to customize or offer alternatives, it can violate locale-choice expectations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The examples set '--tz "America/Vancouver"' and repeat that locale in multiple cron commands, which imposes a specific regional setting in natural-language usage guidance. Because the README does not present this as a placeholder to customize or offer alternatives, it can violate locale-choice expectations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The examples set '--tz "America/Vancouver"' and repeat that locale in multiple cron commands, which imposes a specific regional setting in natural-language usage guidance. Because the README does not present this as a placeholder to customize or offer alternatives, it can violate locale-choice expectations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The template instructs the agent to update queue and memory files as part of normal operation but does not warn users that the agent may autonomously modify local state. That creates a transparency and integrity risk: users may not expect background file edits, and the agent may overwrite or append data without explicit acknowledgment.

Content

No source excerpt is available for this finding.