The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/build.py script utilizes subprocess calls to execute git commands for environment discovery, such as identifying the repository root and managing worktrees. These calls are restricted to local repository metadata and serve a necessary functional role in resolving relative evidence paths.
[DATA_EXFILTRATION]: The skill implements proactive measures to prevent sensitive data leakage. The build script contains an extensive suite of regular expressions to detect and block credentials, tokens, and authorization headers from being included in the generated evidence. Furthermore, it supports pixel-burning redaction for screenshots, ensuring sensitive visual information is permanently removed before the artifact is created.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted JSON evidence packs, presenting a data ingestion surface. This risk is managed through a documented evidence chain: ingestion occurs in scripts/build.py; boundary markers and masking rules are defined in references/privacy.md; the capability inventory is limited to local file writing and network-less rendering; and sanitization is enforced through regex-based secret scanning and HTML escaping.

evidence-artifact