The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from the user's project files to identify compliance signals, creating a surface for potential injection from untrusted data within the repository.
Ingestion points: The scripts check-isaac.mjs and isaac-audit.mjs read up to 600 and 350 local text files, respectively.
Boundary markers: Absent; the tool concatenates file contents for string-based analysis.
Capability inventory: Local file system read/write access and Git hook installation.
Sanitization: Absent; the scripts perform keyword matching on raw file content.
[PERSISTENCE]: The skill provides scripts/install-isaac-hooks.sh, which installs a Git hook to automate Isaac checks during the development workflow. This is a transparent and user-initiated persistence mechanism.
[REMOTE_CODE_EXECUTION]: Installation instructions refer to downloading the skill from the vendor's repository on GitHub (github.com/wundercorp/isaac-skill). This is a standard installation pattern targeting the official vendor source.
[DATA_EXFILTRATION]: Audit scripts scan for sensitive strings such as API keys and analytics SDKs to ensure they are documented for App Store review. No code was found that attempts to transmit this data to external servers.

isaac