The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code diffs to generate narration. Maliciously crafted code comments or content within the diff could attempt to influence the agent's behavior or narration output.
Ingestion points: The orchestrator reads the contents of diff.txt (a git diff) to cluster topics, and workers read individual hunks to provide narration.
Boundary markers: The skill does not define explicit delimiters or instructions to ignore embedded commands within the diff content itself.
Capability inventory: The agents have access to subprocess.run (via the helper script) for command execution and the Read/Write tools for file access.
Sanitization: No specific sanitization or filtering is performed on the diff content before it is processed by the agent reasoning steps.
[COMMAND_EXECUTION]: The helper script bin/difftour.py executes git, gh, and glab commands to resolve diff ranges and fetch pull request information. These calls represent a capability tier used for local repository access.
Evidence: The run_git and git_out functions in bin/difftour.py invoke CLI tools via subprocess.run. These calls use list-based arguments which mitigate direct shell injection vulnerabilities, but they grant the skill the ability to perform network operations (git fetch) and read repository metadata.

diff-tour