The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill employs the !command syntax within its script directory (specifically in scripts/create-msw-handler.md, scripts/create-test-case.md, and scripts/create-test-fixture.md). This syntax triggers shell command execution at the moment the skill is loaded, replacing placeholders with the command output.
[CREDENTIALS_UNSAFE]: The file scripts/create-msw-handler.md contains a dynamic context command: grep -r "API_URL\|VITE_API\|NEXT_PUBLIC_API" .env*. This command recursively scans files matching .env*, which are high-value targets containing sensitive environment variables, secrets, and credentials. Accessing these files via automated load-time execution is an unsafe practice.
[COMMAND_EXECUTION]: The skill executes multiple shell commands (grep, find, wc, date, python) across its metadata and helper scripts to perform environment detection. This bypasses the typical manual tool-invocation review process for these specific discovery tasks.
[INDIRECT_PROMPT_INJECTION]: The skill's scaffolding scripts ingest user-provided $ARGUMENTS directly into the generated file templates. There is a lack of boundary markers or sanitization to ensure that these inputs do not contain malicious instructions that could manipulate the agent's behavior during the file creation or testing process.

testing-unit