The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/scout-analyze.sh contains commands to install system packages using sudo apt-get install -y. This represents an attempt to escalate privileges and modify the host system configuration.\n- [EXTERNAL_DOWNLOADS]: The script attempts to download and install external software from an unverified source using go install github.com/boyter/scc/v3@latest. This bypasses standard package management and introduces unverified code into the environment.\n- [REMOTE_CODE_EXECUTION]: By downloading source code from a remote repository and installing it as an executable binary, the skill facilitates a remote code execution vector.\n- [DATA_EXFILTRATION]: The analysis script explicitly searches for sensitive configuration files such as .env.local and .env.example. Accessing these paths is a key component of data exposure and potential exfiltration, especially when combined with git history analysis that extracts contributor and commit metadata.\n- [PROMPT_INJECTION]: The skill's primary function is to ingest and summarize large amounts of untrusted codebase data. This creates a significant surface for indirect prompt injection. Ingestion points: scripts/scout-analyze.sh (find/grep) and the Gemini-based large codebase strategy described in SKILL.md. Capabilities: The skill has access to the Bash tool for subprocess execution and writes to the filesystem (plans/ folder). Sanitization: No boundary markers or sanitization logic were detected to prevent the agent from obeying instructions hidden within the analyzed files.

dev-scout