The Agent Skills Directory

[SAFE]: The skill is entirely defensive in nature, providing a 'SOTA' (State of the Art) playbook for dependency management of AI instructions. It correctly identifies and provides mitigations for supply chain risks in the agent ecosystem.
[PROMPT_INJECTION]: Static detectors flagged instructions such as 'ignore previous instructions' and 'do not tell the user' in the file rules/02-trust-boundary-and-capability.md. Technical review confirms these strings are components of a git log grep command intended to help agents find evidence of past injections in a repository's history, rather than being active injection attempts themselves.
[COMMAND_EXECUTION]: The skill provides several shell command examples (e.g., git ls-files, grep, find) intended for use during a security audit. These commands are read-only and designed to help the agent or user inventory loaded instruction files and identify self-granted tool permissions. They do not execute unauthorized or harmful code.
[DYNAMIC_CONTEXT_INJECTION]: The skill includes a detailed analysis of the ! shell execution syntax (dynamic context injection) found in some agent platforms. It documents how this feature can be misused for silent command execution and provides clear instructions on how to disable it (disableSkillShellExecution: true) and audit for its presence in third-party files.
[EXTERNAL_DOWNLOADS]: References to external resources point exclusively to official documentation at code.claude.com, which is a well-known and trusted service domain belonging to the platform vendor.

sota-skill-security