The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions that direct the agent to prioritize the skill's current content over its own memory or past conversation history ("スキルは作業を始めるたびに入れ直す。前回読んだ内容やメモリで済ませない", "メモリや過去の会話の手順がスキルと食い違ったら、スキルに従う"). This follows the pattern of overriding previous instructions to maintain context control.
[INDIRECT_PROMPT_INJECTION]: The skill directs the agent to read and follow external repository configurations (conventions.yml) and user-provided Issue/PR comments to guide implementation and metadata formatting.
Ingestion points: Instructions to read .github/conventions.yml and Issue/PR comments within the SKILL.md file.
Boundary markers: There are no defined delimiters to separate external data from instructions.
Capability inventory: The skill allows the agent to modify files, perform git operations (commit, push), and use gh CLI to manage Issues and Pull Requests.
Sanitization: The skill does not specify any sanitization or validation steps for data read from the repository or GitHub API.
[COMMAND_EXECUTION]: The skill utilizes shell commands to inspect environment variables (env | grep), configure local Git behavior (git config core.hooksPath), and manage development tools (lefthook install).
[DYNAMIC_EXECUTION]: The skill instructs the agent to implement or modify Git hooks (specifically to remove Co-authored-by fields) which involves generating and placing executable scripts into the local repository environment.

github