Back to skill

Security audit

Agent Task Manager

Security checks for vulnerabilities and agentic risk

Overview

The skill’s external API use is mostly disclosed, but it needs Review because it sends free-form task data off-device, can trigger outbound notifications, and includes an unsafe shell wrapper.

Install only if you are comfortable with task prompts, workflow details, and notification content being processed by the SkillBoss API. Do not include secrets or regulated data in task requests, keep SKILLBOSS_API_KEY in protected environment configuration, and avoid using cooldown.sh with any untrusted or generated command/task-name input until the eval and path handling are fixed.

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 (4)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/cooldown.sh:9
Finding

Shell Command Injection Through eval-Based Command Execution

Content
View full analysis
"$TIMESTAMP_FILE" echo "✅ Success. Timestamp updated." else echo "❌ Command failed. Timestamp NOT updated." exit 1 fi ``` ### Technical Analysis All command arguments are flattened into the `COMMAND` string and subsequently passed to `eval`. Unlike direct argument execution, `eval` causes shell metacharacters, substitutions, redirections, and command separators contained in the input to be interpreted a second time. The wrapper is intended to execute a supplied command, but `eval` exceeds what is necessary for that functionality. It destroys argument boundaries and allows data intended to be an ordinary argument to become shell syntax. This becomes exploitable whenever a trusted caller invokes the wrapper with a command containing partially untrusted values. For example, a value containing `;`, `$(...)`, backticks, or shell redirection can execute an additional command under the account running the Skill. ### Attack Path 1. An attacker supplies or influences an argument passed to `cooldown.sh`. 2. A trusted caller invokes the wrapper, for example: ```bash ./scripts/cooldown.sh monitor 60 echo 'untrusted; id > /tmp/injected' ``` 3. The script joins the command and its arguments into one string. 4. `eval` reparses that string as shell source. 5. The injected command executes with the privileges and environment of the Skill process. 6. Depending on process permissions, the attacker can read or modify accessible files, invoke network utilities, a ...[truncated 571 chars]
Remediation
View remediation
"$TIMESTAMP_FILE" else exit 1 fi ``` - Require at least one command argument and reject an empty command. - If only a fixed set of operations is needed, replace generic command execution with an explicit allowlist. - Do not build shell commands by concatenating untrusted values. - Run the wrapper under a minimally privileged service account without access to unrelated credentials or sensitive files. - Add tests using arguments containing spaces, semicolons, command substitutions, glob characters, and redirections to verify that they remain literal arguments. ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/cooldown.sh:9
Finding

Arbitrary File Write and File Corruption Through Unsanitized Task Name

Content
View full analysis
"$TIMESTAMP_FILE" ``` ### Technical Analysis `TASK_NAME` is incorporated directly into a filesystem path without validation. A value containing directory traversal components such as `../` can escape `agent_task_manager_data`. After a wrapped command succeeds, the resulting path is overwritten with a timestamp. The fixed `_last_run.txt` suffix limits which final names can be selected, but does not prevent escaping the intended directory or corrupting an existing writable file whose name matches the resulting suffix. The timestamp directory is relative to the current working directory, which also makes the write destination dependent on how the script is launched. ### Attack Path 1. An attacker controls or influences the task name supplied as the first argument. 2. The attacker supplies traversal components, such as: ```bash ./scripts/cooldown.sh "../../tmp/target" 0 true ``` 3. The script constructs a path resembling: ```text ./agent_task_manager_data/../../tmp/target_last_run.txt ``` 4. The wrapped command succeeds. 5. The script follows the traversed path and overwrites the destination with a timestamp. 6. If the process can write to the selected location, an existing file is corrupted or a new file is created outside the intended state directory. Symlinks placed at a predictable timestamp-file path could also redirect the write to another writable target. ### Impact Assessment An attacker can create or overwrite timestamp-suffixed files reachable with the Skill process's files ...[truncated 355 chars]
Remediation
View remediation
&2 exit 1 fi ``` - Resolve and verify that the canonical timestamp path remains beneath the canonical state directory. - Use a fixed state directory independent of the caller's working directory. - Create the directory with restrictive permissions, such as `0700`. - Defend against symlink attacks by using safe file-opening techniques or by checking that the target is not a symlink immediately before writing. - Use atomic state updates through a securely created temporary file followed by a rename. - Store task-name-to-file mappings using a safe encoding or cryptographic hash rather than embedding raw identifiers in paths. ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/task_parser.py:58
Finding

Unredacted Human Requests Disclosed to an External LLM Service

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/molt_task.py:26
Finding

Workflow State and Results Persisted in Plaintext with Predictable Permissions

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
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (19)

Tainted flow: 'SKILLBOSS_API_KEY' from os.environ (line 10, credential/environment) → requests.post (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.

Content

Scanner excerpt · scripts/orchestrator.py (reported line 15)May include surrounding context.

python
def _pilot(body: dict) -> dict:
    r = requests.post(
        f"{API_BASE}/pilot",
        headers={"Authorization": f"Bearer {SKILLBOSS_API_KEY}", "Content-Type": "application/json"},
        json=body,

Tainted flow: 'SKILLBOSS_API_KEY' from os.environ (line 7, credential/environment) → requests.post (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.

Content

Scanner excerpt · scripts/task_parser.py (reported line 12)May include surrounding context.

python
def _pilot(body: dict) -> dict:
    r = requests.post(
        f"{API_BASE}/pilot",
        headers={"Authorization": f"Bearer {SKILLBOSS_API_KEY}", "Content-Type": "application/json"},
        json=body,

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The skill claims stateful orchestration, error recovery, and rate-limit management, but the documented behavior suggests a narrower parsing/translation role with external API dependence. This is dangerous because operators may rely on security or safety properties that are not actually implemented, leading to unsafe automation, unintended retries, or data handling assumptions that fail in production.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The skill claims stateful orchestration, error recovery, and rate-limit management, but the documented behavior suggests a narrower parsing/translation role with external API dependence. This is dangerous because operators may rely on security or safety properties that are not actually implemented, leading to unsafe automation, unintended retries, or data handling assumptions that fail in production.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
93% confidence
Finding

The skill declares environment, file, and network-relevant behavior but does not declare any explicit tool scope or permissions boundaries. In a workflow/orchestration skill, this is dangerous because it can cause the agent to invoke sensitive capabilities implicitly, including reading state files, writing persistent task data, and sending data to external services without clear user consent or policy gating.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The invocation language is broad enough to match common requests for creating workflows or improving sequential tasks, which can cause the skill to be selected in contexts where users did not intend external API use or persistent state operations. In this skill's context, broad matching increases the chance of silent data transmission and overreach into general agent activity.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The documentation states that AI inference and notification delivery are handled through an external API hub, but it does not clearly warn users that task data may leave the local environment. This is dangerous because workflow contents can include sensitive operational, financial, or personal information that may be transmitted to a third party without informed consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The example workflow includes email/SMS-style alerting but does not clearly warn that the skill can initiate outbound communications. This is dangerous because users may trigger real-world notifications to third parties unintentionally, creating privacy, spam, or social-engineering risks.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
88% confidence
Finding

The skill explicitly references an external API endpoint for both LLM inference and notification delivery, meaning user inputs and workflow data may be transmitted outside the local trust boundary. In the context of a task manager that may handle rich state and dependency data, this external transmission materially increases confidentiality and compliance risk.

Content

Scanner excerpt · SKILL.md (reported line 46)May include surrounding context.

md
All AI inference and notification delivery is handled through **SkillBoss API Hub**:

- **Endpoint:** `https://api.heybossai.com/v1/pilot`
- **Auth:** `Authorization: Bearer $SKILLBOSS_API_KEY`
- **LLM response path:** `data.result.choices[0].message.content`

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The code loads a third-party API key and sends workflow content to a remote vendor for chat/email operations, but that external-service dependence is not justified by the stated task-manager role. In context, this is more dangerous because a workflow orchestrator has broad access to user tasks and state, so hidden remote processing can cause unexpected disclosure of sensitive workflow data.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
84% confidence
Finding

This code posts arbitrary request bodies to an external API, creating a data egress path from the skill to a third party. In this skill's context, that matters because parsed tasks, workflow content, and notification data may be routed off-system without strong transparency or minimization controls.

Content

Scanner excerpt · scripts/orchestrator.py (reported line 15)May include surrounding context.

python
def _pilot(body: dict) -> dict:
    r = requests.post(
        f"{API_BASE}/pilot",
        headers={"Authorization": f"Bearer {SKILLBOSS_API_KEY}", "Content-Type": "application/json"},
        json=body,

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill manifest describes a generic workflow/task manager, but the implementation embeds domain-specific financial monitoring and outbound notification behavior. This capability mismatch is dangerous because users or reviewers may approve the skill expecting orchestration only, while it can process financial prompts and trigger external actions that expand data-handling and abuse risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The notification path transmits recipient and message content to an external API without any visible consent, confirmation, or policy check. In a task-manager context, this is risky because workflow state may contain user-generated or sensitive content, and the code can send it externally as part of an automated step.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill sends user-provided natural-language requests to a third-party LLM service, but that data-sharing behavior is not clearly justified or bounded in the code. In a task-management skill, requests may contain operational plans, credentials, internal identifiers, or other sensitive workflow context, so undisclosed off-platform transmission increases privacy and supply-chain risk.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · scripts/orchestrator.py (reported line 11)May include surrounding context.

python
import requests

SKILLBOSS_API_KEY = os.environ["SKILLBOSS_API_KEY"]
API_BASE = "https://api.heybossai.com/v1"


def _pilot(body: dict) -> dict:

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · scripts/task_parser.py (reported line 8)May include surrounding context.

python
import requests

SKILLBOSS_API_KEY = os.environ["SKILLBOSS_API_KEY"]
API_BASE = "https://api.heybossai.com/v1"


def _pilot(body: dict) -> dict:

External Transmission

Medium
Category
Data Exfiltration
Confidence
80% confidence
Finding

This code performs external network transmission to a third-party endpoint as part of request parsing. In context, the danger is not the mere presence of HTTP but that arbitrary user task descriptions are exported off-system, which can expose sensitive workflow data and make the agent dependent on an external service for core behavior.

Content

Scanner excerpt · scripts/task_parser.py (reported line 12)May include surrounding context.

python
def _pilot(body: dict) -> dict:
    r = requests.post(
        f"{API_BASE}/pilot",
        headers={"Authorization": f"Bearer {SKILLBOSS_API_KEY}", "Content-Type": "application/json"},
        json=body,

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

Unknown requests are forwarded verbatim to an external API, and this file provides no warning, consent flow, or filtering for sensitive content. Because this parser is meant to process arbitrary human requests, it is reasonably likely to receive secrets, personal data, or internal business instructions that should not leave the environment by default.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

Requiring an API key without warning about secure handling of the credential can lead users to expose or mishandle secrets in shared environments, logs, or task state. While this issue is less severe than direct exfiltration, it still weakens operational security around a credential used for external API access.

Content

No source excerpt is available for this finding.