The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill explicitly directs the agent to access a 'Secrets vault' located at a relative path '../secrets/' outside the project directory. It instructs the agent to search for and utilize Personal Access Tokens (PAT), SSH keys, or API credentials stored in this location for automation purposes, which increases the risk of sensitive data exposure.
[COMMAND_EXECUTION]: The workflow involves executing a PowerShell script ('../agent-skills/scripts/sync-faruk-skills.ps1') from a sibling directory to synchronize environment settings. Executing scripts from paths outside the controlled project environment is a security concern as it relies on the integrity of external files that are not part of the skill itself.
[EXTERNAL_DOWNLOADS]: The skill uses 'npx skills add' and 'npx skills experimental_install' to download and install agent skills from remote repositories, including 'obra/superpowers' and 'farukzahra/agent-skills'. The use of the '-y' flag to automatically install all skills in a package further reduces user oversight of the downloaded content.
[INDIRECT_PROMPT_INJECTION]: The initialization process involves reading several existing project files to determine the tech stack, providing a surface for malicious instructions to influence the agent's behavior.
Ingestion points: 'README.md', 'AGENTS.md', 'package.json', and 'docs/stack.md' are read during the 'Inspect' phase.
Boundary markers: No delimiters or specific instructions are provided to the agent to disregard potential natural language instructions embedded within these files.
Capability inventory: The skill has the ability to execute PowerShell scripts, write files to the project directory, and install external software via npx.
Sanitization: There is no evidence of sanitization or validation performed on the content ingested from the project files before it is processed by the agent.

project-init