The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions and its reference document references/analysis-checklist.md direct the agent to actively search for and read sensitive configuration files, specifically targeting .env files and code patterns involving environment variables (e.g., os.environ). This exposes system secrets to the AI context as part of its documentation workflow.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted third-party codebases through tools like Read, Grep, and Glob. This ingestion creates a vulnerability where an attacker could embed malicious instructions in code comments or strings within a project to influence the agent's actions. Ingestion points: Project source files processed via Read and Grep. Boundary markers: None identified; there are no instructions for the agent to ignore directives found within the analyzed code. Capability inventory: The skill uses Bash, Read, Glob, and Grep tools, providing significant file system access and execution power. Sanitization: None identified.
[COMMAND_EXECUTION]: The skill requests and utilizes the Bash tool for codebase exploration and mapping. While intended for discovery, the presence of a general-purpose shell tool alongside instructions to access sensitive files increases the potential for misuse.

spec-miner