The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to interpret "intent" from external, potentially user-controlled data sources such as git logs, commit messages, and kanban task summaries.
Ingestion points: Git log output, commit messages from branch history, and external task summaries via the hermes kanban show command.
Capability inventory: The agent has permissions to write to the filesystem, execute shell commands via terminal (including git and build scripts), and update task metadata.
Boundary markers: The instructions do not define strict delimiters or safety warnings for the agent when processing these external text blocks as intent instructions.
Sanitization: No explicit filtering or validation steps are defined to prevent the agent from acting on malicious instructions embedded in commit messages.
[DYNAMIC_EXECUTION]: The skill requires the execution of arbitrary code within the repository to verify that the merge resolution is correct.
The procedure directs the agent to "Run the project's build/tests via terminal" and to "import/execute the touched modules."
Executing code that has been modified during a conflict (and potentially by multiple parties) presents a risk where malicious code could be executed during the verification phase.
[COMMAND_EXECUTION]: The skill relies on terminal access to perform various git operations and run project scripts.
Executed commands include git status, git merge-base, git log, git diff, git add, and git commit alongside user-defined build and test commands.

agent-merge-conflict-arbiter