The Agent Skills Directory

[Instruction Generation from Session Data]: The skill facilitates the creation and modification of agent instructions based on session learnings. While this is its primary purpose, there is a security consideration regarding indirect prompt injection. If the agent processes malicious data (such as code or pull request comments) and identifies it as a 'learning', that malicious content could be persisted into the repository's instructions, potentially influencing future agent sessions.
Evidence: The procedure involves identifying learnings from the session and writing them to .github/skills/ or .github/instructions/.
[Shell Command Execution]: The skill utilizes standard shell commands to manage and search for existing repository knowledge. These operations are limited to listing and searching within specific, expected directories.
Evidence: Usage of ls .github/skills/, ls .github/instructions/, and grep -r to identify existing content.

update-skills