The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted data from the Git environment that could contain malicious instructions for the agent.\n
Ingestion points: Data from git log, git diff, git status, and git branch (including branch names, commit messages, and file contents) are read from the local repository and its remotes into the agent's context.\n
Boundary markers: The skill lacks explicit delimiters or instructions to the agent to disregard potential commands embedded within the Git output, which could lead to the agent following instructions found in commit messages.\n
Capability inventory: The agent utilizes the Bash tool and has the capability to modify the local filesystem via commands like git merge and git rebase.\n
Sanitization: No sanitization, filtering, or validation is performed on the data retrieved from Git commands before it is processed or presented to the agent.\n- [COMMAND_EXECUTION]: The skill executes standard Git commands via Bash to manage branch state and resolve conflicts. These operations are restricted to the local repository and are essential to the primary purpose of the skill.\n- [EXTERNAL_DOWNLOADS]: The skill performs network operations using git fetch to retrieve updates from remote repositories. This behavior is expected for a Git synchronization tool and targets the user's configured remotes.

git-reconciler