The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/check_commit_msg.py executes git log via subprocess.run to audit commit ranges. The implementation uses a list-based argument structure and avoids shell invocation (shell=False), which effectively prevents command injection from the user-supplied revision range argument.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted commit messages via scripts/check_commit_msg.py from multiple ingestion points (message files via --file, stdin, and git log history via --range). While it lacks explicit natural language boundary markers, it employs structural validation via regular expressions. The skill's capability inventory is limited to executing git log and reading local files, and sanitization is performed by enforcing strict formatting rules for the subject line. These factors significantly mitigate the risk of indirect prompt injection.\n- [SAFE]: The inclusion of the Variation Selector-16 (U+FE0F) Unicode character in scripts/check_commit_msg.py is a standard emoji handling practice for normalization and is not used as an obfuscation technique.\n- [SAFE]: Behavioral instructions in SKILL.md, such as output constraints ("Output only the commit message") and persistence of style, are functional requirements for the skill's purpose and do not represent attempts to bypass agent safety guidelines or ethical protocols.

gitmoji-conventional