The Agent Skills Directory

[PERSISTENCE]: The skill provides instructions for the agent to modify user shell configuration files. Specifically, in 'references/day1_onboarding.md', under 'Problem 3: CLI command not found', it instructs adding export commands to ~/.bashrc or ~/.zshrc to persistently alter the PATH environment variable. In an agent context, instructions to modify startup scripts are considered high-risk persistence mechanisms.
[METADATA_POISONING]: The skill includes a file named .security-scan-passed which contains a 'Security scan passed' message, a timestamp, and a content hash. This mimics the output of an automated security tool to deceive users or auditors into believing the skill has been pre-verified, which is a common metadata poisoning technique.
[INDIRECT_PROMPT_INJECTION]: The skill implements an 'Autonomous Execution' framework (detailed in 'references/master_qa_prompt.md') that is vulnerable to indirect prompt injection.
Ingestion points: The agent is instructed to read external, potentially untrusted test case documents (e.g., 02-CLI-TEST-CASES.md).
Boundary markers: None are present; there are no instructions to disregard potentially malicious commands embedded in the test cases.
Capability inventory: The agent is granted the capability to execute shell commands ('Execute all steps exactly as documented') and write to the filesystem (via provided Python scripts).
Sanitization: None. The instructions prioritize absolute compliance with the external document's steps over safety filters.
[COMMAND_EXECUTION]: The skill encourages the use of a 'Master Prompt' that instructs the agent to execute shell commands derived from test cases. For example, 'references/day1_onboarding.md' and 'assets/templates/TEST-CASE-TEMPLATE.md' provide examples of complex shell operations, including git clone, pnpm install, docker exec, and rm -rf operations, which the agent is directed to perform autonomously.

software-testing-guide