The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to search for and read sensitive GitHub Personal Access Tokens (PATs) and environment variables from hardcoded local paths, including C:\faruk\secrets\pat.txt, C:\repo\secrets\github\pat.txt, and .env files. This pattern involves searching for credentials in sensitive storage locations.
[EXTERNAL_DOWNLOADS]: The skill installs external agent extensions using the npx skills add command from repositories not identified as trusted organizations, specifically obra/superpowers and mattpocock/skills@handoff. While the author's own skill farukzahra/agent-skills is also installed, the inclusion of third-party sources introduces supply chain risks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests raw, unsanitized user prompts and stores them verbatim in a PLAN.md file. This content is then used as instructions for subsequent automated planning and execution skills (brainstorming, writing-plans). This creates a vulnerability surface where a user-provided prompt could contain instructions that manipulate the agent's behavior during those automated phases.
[COMMAND_EXECUTION]: The skill performs global system modifications by updating the global git configuration (git config --global). It also executes complex PowerShell commands to interact with the GitHub API and performs authenticated git operations using dynamically constructed headers.
[DYNAMIC_EXECUTION]: The skill uses runtime logic to convert credentials into base64 encoded strings for HTTP Basic authentication headers ([Convert]::ToBase64String) and injects these into shell commands to facilitate git pushes.

init-fmz