The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is designed to capture and display the entire environment state, including all environment variables. As demonstrated in the provided reference example (worked-example-windows.md), this process captures sensitive credentials such as DEEPSEEK_API_KEY and GITHUB_TOKEN if they are present in the system environment. The instructions explicitly direct the agent to dump all environment variables to compare the state before and after activation.
[COMMAND_EXECUTION]: The skill executes a wide range of local CLI tools (e.g., compilers, runtimes, package managers) to determine their versions and paths. The probe-devtools.ps1 and probe-devtools.sh scripts iterate through a hardcoded list of dozens of tools, executing each with version-check flags (e.g., --version, -V). While this is the intended functionality, it involves widespread subprocess execution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes output from local CLI tools and environment variable values. Every ingestion point is a potential surface where a malicious tool or a compromised environment variable could inject instructions into the agent's context. Although the scripts implement timeouts and basic filtering to avoid hanging or capturing excessive output, the boundary between data and instructions remains porous.

dev-env-inventory