T09 · Insecure Skill Coding Practices
- Location
SKILL.md:13- Finding
Execution of Unbundled and Unverified External Scripts
- Content
View full analysis
Vulnerability Details
File Location:
SKILL.md, lines 13-22
Vulnerability Type: Execution of externally mutable, unaudited local scripts
Risk Level: MediumVulnerable Code
markdown ## Includes - gru_yolo_scout.py - gru_bounty_closer.py - gru_moltbook_guard.py ## Run (from raddonpipeline scripts) ```bash cd ~/.hermes/profiles/raddonpipeline/scripts python3 gru_yolo_scout.py && python3 gru_bounty_closer.pytext ### Technical Analysis The documented command changes into `~/.hermes/profiles/raddonpipeline/scripts`, a user-profile directory outside the audited project, and executes two Python files by relative name. None of the three advertised Python files is included in the project, whose only audited file is `SKILL.md`. Consequently, the implementation, provenance, integrity, and effective behavior of the executed code cannot be verified from the Skill package. Because the external directory is independently mutable, any process or user able to create or replace the named files can control what is executed when these instructions are followed. The scripts execute with the permissions of the invoking user or Agent. The advertised `gru_moltbook_guard.py` is also not invoked by the documented command, so the claimed suspension, rate-limit, and deduplication protections are not demonstrably enforced. ### Attack Path 1. An attacker or compromised process obtains write access to `~/.hermes/profiles/raddonpipeline/scripts`. 2. The attacker creates or replaces `gru_yolo_scout.py` or `gru_bounty_closer.py` with attacker-controlled Python code. 3. A user or Agent follows the instructions in `SKILL.md`. 4. The shell enters the externally mutable directory and Python executes the attacker-controlled file. 5. The payload operates with the invoking account's permissions and can access resources available to that account. This path is conditional on an attacker being able to modify the e ...[truncated 740 chars]- Remediation
View remediation
Remediation Suggestions
- Include every required Python script in the reviewed Skill package rather than relying on files from a user-profile directory.
- Resolve executable paths relative to the trusted Skill installation directory and reject paths that escape that directory.
- Verify packaged scripts with cryptographic hashes or signed release artifacts before execution.
- Integrate the guard logic into the mandatory execution path instead of merely listing
gru_moltbook_guard.py. - Fail closed when suspension state, comment limits, or deduplication checks cannot be loaded or validated.
- Run the scripts with least privilege, a restricted environment, minimal filesystem access, and only the network and credentials required for the stated task.
- Document and pin all runtime dependencies so the complete execution chain can be audited reproducibly.
