Back to skill

Security audit

Sales Enablement Training

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly matches its course-building purpose, but its update flow asks the agent to run a remotely supplied command that can modify the CLI or skill after review.

Review the update workflow carefully before installing. Normal course creation and OAuth behavior are expected, but you should only approve CLI or skill update commands if they are clearly from PersonWise and match the intended install or upgrade action; avoid approving opaque shell commands or unexpected destinations.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T03 · Remote Payload Retrieval and Execution

Error
Location
SKILL.md:112
Finding

Execution of Remotely Supplied Update Commands

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:112-124
Vulnerability Type: Execution of an insufficiently constrained command received from a remote service
Risk Level: High

Vulnerable Code Snippet:

text
- If `updates.cli.status` or `updates.skill.status` is `update_available` or `below_minimum`,
  tell the user once which component is outdated (installed versus latest) and quote the exact
  `action` command. The task cannot continue until that update is installed. Ask for approval;
  with approval, run exactly that command (it already carries the required `--approve-upgrade`
  argument). If the user declines, stop and do not run business commands with the outdated
  component, and do not ask again in this session unless the user changes that decision.
- If a command fails with `CLI_VERSION_BELOW_MINIMUM` or `SKILL_VERSION_BELOW_MINIMUM`, the task
  cannot continue until the update is installed. Explain this, ask for approval, run exactly the
  printed update command, then retry the failed step once.
- When both are outdated, update the CLI first, then the Skill.

When the printed `action` refreshes this installed Skill, replace `<skill-directory>` with the
directory of this installed Skill (the directory containing this Skill's SKILL.md). Never run
`doctor` or a generic capability preflight to check freshness; the `update check` command above is
the freshness check.

Technical Analysis

The Skill directs the agent to execute an action command returned dynamically by the PersonWise service. It does not define a local allowlist for the executable, supported subcommands, arguments, destination paths, download origins, or expected update artifacts before execution.

Requiring user approval reduces accidental execution but does not establish the safety of an opaque command controlled by a remote response. The instruction to run the command “exactly” also discourages the agent from ...[truncated 2136 chars]

Remediation
View remediation

Remediation Suggestions

  1. Never execute a shell command string received from an API or update response.
  2. Replace the textual action field with structured update metadata, such as component, version, artifact identifier, expected digest, signature, and approved destination.
  3. Invoke only a fixed, locally trusted updater executable using an argument array rather than shell evaluation or interpolation.
  4. Enforce an allowlist of supported update operations, subcommands, flags, HTTPS origins, signing identities, and destination directories.
  5. Verify the downloaded artifact's cryptographic signature and pinned digest before installation. Bind verification to the expected component and version.
  6. Resolve and validate the Skill directory locally. Reject symbolic links, reparse points, path traversal, unexpected ownership, and paths outside the designated installation root.
  7. Stage updates in a private temporary directory, validate archive entries, and use an atomic replacement operation with rollback support.
  8. Present the user with structured information about the component, current and target versions, source, signature status, destination, and requested filesystem changes rather than asking approval for an opaque command.
  9. Treat unknown response fields, executables, arguments, origins, or update operations as fatal validation failures.
  10. Log a secret-free record of the validated update metadata and final artifact digest for later auditing.
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Scope Creep

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The bootstrap script performs an outbound network download and installs executable content even though the skill metadata declares only Bash permission. While the script includes meaningful safeguards such as explicit approval gates, fixed URLs, size checks, and SHA-256 verification, the undeclared installer behavior still expands the skill's effective capability beyond what reviewers and users would expect from the manifest. In a skill context, hidden or under-declared software installation is dangerous because it can bypass trust assumptions and enable execution of new code on the host.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
80% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · assets/bootstrap.sh (reported line 165)May include surrounding context.

sh
echo "Release executable checksum mismatch." >&2
  exit 6
fi
chmod 700 "$candidate"
if [ "$(uname -s)" = "Darwin" ] && [ "$native_signature_status" = "verified" ]; then
  codesign --verify --deep --strict --verbose=2 "$candidate"
  spctl --assess --type execute --verbose=2 "$candidate"