The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses the $ARGUMENTS variable to incorporate user input directly into its reasoning context. This creates a surface where a user could provide text that attempts to override the skill's instructions or logic.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes the output of git diff. If a repository contains maliciously crafted code comments or content designed to trigger the LLM, it could influence the poetic output or the final commit message summary.\n
Ingestion points: Reads git diff --cached and git diff output in SKILL.md.\n
Boundary markers: Absent; the diff content is analyzed without delimiters or specific instructions to ignore embedded commands.\n
Capability inventory: Uses the Bash tool to execute git diff and git commit commands across SKILL.md.\n
Sanitization: Absent; the skill does not filter or escape the content of the diff or user arguments before processing them.- [COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform repository operations. While it includes a user confirmation step, the generated commit message—which could be influenced by injected content—is passed into the git commit -m command. The skill relies on the underlying platform's tool-calling implementation to prevent shell injection, but the instruction itself links untrusted data to command execution.

commit-poet