The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill attempts to execute an external Python script using a hardcoded absolute path: /Users/gurusharan/.codex/skills/.system/skill-creator/scripts/quick_validate.py. This script is not included in the skill distribution, and executing files from absolute paths on a local system is a security risk as the code content is unverifiable and environment-dependent.
[COMMAND_EXECUTION]: The scripts/audit.sh script executes various shell commands (find, grep, awk, perl) using the skill_path argument provided by the user. While variables are quoted, the complexity of the script and the lack of rigorous input validation on the directory structure or file names being processed could lead to unexpected behavior or shell command injection if malicious directory structures are audited.
[PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data (the contents of other skills' SKILL.md files).
Ingestion points: The audit.sh script reads and parses the content of SKILL.md from the provided skill_path.
Boundary markers: There are no explicit markers or instructions to the agent to ignore instructions embedded within the audited content.
Capability inventory: The skill has the capability to execute shell scripts and Python scripts (audit.sh, quick_validate.py).
Sanitization: There is no sanitization of the audited markdown content before it is processed by regex-based tools. A malicious skill being audited could contain indirect prompt injections designed to manipulate the auditing agent's scorecard or subsequent actions.

skill-quality-gate