The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a set of Bash and Python scripts to automate project inspection, management of git worktrees, and execution of test runners. These operations are part of the core functionality for project verification.
[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of code found within the target repository (e.g., project-specific test suites). This is a primary feature of the verification framework, which is protected by an explicit --trust requirement and the use of isolated git worktrees and temporary directories.
[DATA_EXFILTRATION]: To prevent accidental data exposure, the skill includes a secret redaction mechanism in its control template. It scans environment variables for patterns like 'secret', 'token', or 'key' and redacts their values from logs and reports.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from project files (such as CLAUDE.md, AGENTS.md, and JSON feature records) to configure its execution. It explicitly instructs the agent to treat this content as data rather than instructions to mitigate injection risks.
Ingestion points: CLAUDE.md, AGENTS.md, package.json, and various JSON feature/scenario records.
Boundary markers: SKILL.md contains explicit instructions to treat instructions in untrusted documents and tool output as data.
Capability inventory: Subprocess execution of git, python3, bash, and project-defined test commands (e.g., npm, vitest).
Sanitization: Implements strict path validation to prevent traversal, uses shlex.quote for shell command construction, and utilizes a pattern-based secret redactor.
[DYNAMIC_EXECUTION]: The skill utilizes Python scripts embedded within Bash heredocs to handle complex tasks like process table monitoring, SHA-256 manifest generation, and atomic file publication.

leyline