The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on scripts/reset-attempt.sh, a shell script designed to reset the workspace after a failed repair attempt. The script carefully manages Git snapshots and removes only untracked files added during the attempt. It employs defensive programming techniques, including the use of -- to separate flags from filenames and localized path management within the Git tree.
[INDIRECT_PROMPT_INJECTION]: The workflow involves processing external bug reports and tool outputs.
Ingestion points: Bug reports are parsed in references/reproduce.md to define symptoms and reproduction steps.
Boundary markers: The instructions do not specify explicit delimiters for external content, but they require strict reproduction of exact steps and verbatim error text.
Capability inventory: The skill has the capability to write files, commit code to Git, and execute local scripts such as quality.py.
Sanitization: There is no explicit sanitization of error logs or reports, which is typical for technical debugging tasks. The risk is minimized by the structured nature of the piece system and the requirement for manual validation.
[DYNAMIC_EXECUTION]: The skill instructs the agent to run a local utility quality.py for reachability and strength testing. This script is treated as a trusted local tool for measuring test coverage and code quality within the development environment.

repair