The Agent Skills Directory

Development Command Execution: The skill instructs the agent to use standard software development and testing tools to manage environments and verify code changes. These operations are consistent with the skill's primary purpose of bug triage and fixing.
Evidence includes the use of uv for environment synchronization, git for diffing changes, and pytest and pnpm for running test suites.
These tools are used within the local project context to ensure the integrity of the development workflow.
Indirect Prompt Injection Surface: As a debugging tool, the skill naturally ingests untrusted data from bug reports and external log files. This is a known surface for indirect instructions, and while inherent to the task, it warrants awareness.
Ingestion points: External data enters the context through the $ARGUMENTS placeholder (bug reports) and by reading log files such as driver/result.json, driver/events.jsonl, and grader/report.md.
Boundary markers: The instructions emphasize capturing exact, verbatim code and output to ensure accuracy, though they do not implement specific security delimiters for this data.
Capability inventory: The skill possesses capabilities for shell command execution, environment manipulation via uv and pnpm, and automated testing.
Sanitization: No specific sanitization or filtering of the ingested bug reports is described, as the agent requires the raw data for accurate reproduction and analysis.

amplifier-agent-bugfix