The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install an NPM package named '@openai/codex'. This scope and package name do not appear to be official OpenAI releases, posing a significant supply chain risk through potential typosquatting or malicious unofficial packages.\n- [REMOTE_CODE_EXECUTION]: The skill relies on and executes an external CLI tool ('codex') whose source and integrity cannot be verified, especially given the suspicious installation instructions.\n- [COMMAND_EXECUTION]: The skill executes several binaries located in '/.claude/skills/gstack/bin/' (e.g., 'gstack-update-check', 'gstack-config', 'gstack-telemetry-log'). These scripts are external to the skill and their behavior is opaque.\n- [DATA_EXFILTRATION]: A telemetry system is implemented that collects usage data. While a user prompt is included, the data is processed by an external binary ('gstack-telemetry-log'), and the preamble logs project-specific information like repository names to local files which could be targeted for exfiltration.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. Ingestion points: User-provided focus instructions and project plan files ('/.claude/plans/*.md'). Boundary markers: None. Input is interpolated directly into prompts for the external Codex model. Capability inventory: Uses 'Bash' to execute the 'codex' CLI and 'Write' to modify project plan files. Sanitization: None.

codex