Back to skill

Security audit

Session Handoff

Security checks for vulnerabilities and agentic risk

Overview

The skill matches its handoff purpose, but it can persist detailed session history and modify handoff files without redaction or path-safety checks.

Install only if you are comfortable with session details being saved to project files. Review generated handoffs before committing or sharing them, avoid invoking it after secrets appeared in the session, and check any CURRENT file before letting the skill update an existing handoff.

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 (2)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:97
Finding

Unredacted Session Data May Be Persisted in Handoff Files

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:65
Finding

Unvalidated CURRENT Pointer May Cause Writes Outside the Handoff Directory

Content
View full analysis
` Do not rewrite handoffs already marked `done` or `superseded`. Use `superseded` only when you are explicitly replacing the same work stream. ``` ### Technical Analysis The skill treats the contents of `CURRENT` as a file pointer and directs the agent to modify the referenced file. It does not require the target to be: - A relative path beneath the selected handoff directory - Canonically resolved and checked for directory containment - A regular file rather than a symbolic link - Free from `..` traversal components - Free from absolute paths - A valid handoff document owned by the current project Because repository contents are potentially attacker-controlled, a malicious project can provide a crafted `CURRENT` file. If the agent follows that pointer literally, it may update an unintended file accessible with the agent's existing filesystem permissions. The instructed modification is limited to handoff metadata, but an implementation may rewrite the target document while changing its front matter. Consequently, the exact degree of corruption depends on how the agent performs the update. ### Attack Path 1. An attacker prepares a repository containing a `CURRENT` file. 2. `CURRENT` contains an absolute path, a traversal path such as `../../target.md`, or a path to a symlink that resolves outside the handoff directory. 3. A victim opens the repository with an agent that has filesystem write access and invokes the session-handoff skill. 4. The skill checks the attacker-selected target and interprets it as an existing active handoff. 5. The agent updates the target with `status: paused` and a new `updated_at` val ...[truncated 829 chars]
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 (4)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger phrases are broad, conversational, and overlap with common user language like 'summarize this session' or 'hand this off'. That increases the chance of accidental invocation, which can cause unintended file creation or updates and expose session details into project artifacts without explicit user confirmation.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
84% confidence
Finding

The instruction to 'Do not ask the user questions' forces autonomous inference from conversation history and git state, then proceeds to write files. That removes an important consent and validation checkpoint, increasing the risk of incorrect summaries, unintended persistence of sensitive data, or writes to the wrong handoff stream.

Content

Scanner excerpt · SKILL.md (reported line 29)May include surrounding context.

md
Use this skill when the user says "handoff", "summarize this session", "create a handoff", "hand this off", or invokes `/handoff`.

Produce a precise handoff document and write it to a file. Do not ask the user questions — derive everything from conversation history and git state.

If the user provides arguments, treat them as the intended focus of the next session and prioritize that focus in **Goal**, **Current State**, and **Next Steps**.
If the user says "update handoff" or similar, update the most recent handoff file in place instead of creating a new one.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill is instructed to write artifacts and may overwrite the most recent handoff in place, but it provides no warning or confirmation before modifying repository files. In practice this can silently alter tracked documentation, clobber an existing handoff, or create persistence for sensitive session content the user did not intend to save.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 261)May include surrounding context.

md
## Anti-patterns

- Do not ask the user "what did we work on?" — derive from conversation.
- Do not write vague entries like "various files updated" — be specific.
- Do not skip the git commands — they provide objective ground truth.
- Do not create a new handoff if the user says "update handoff" — overwrite the most recent one.