The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute local commands for debugging purposes. This includes running a provided shell script template (scripts/hitl-loop.template.sh), triggering unit/integration tests, and using CLI tools like git bisect. These operations are limited to the local development environment for the purpose of reproducing reported bugs.
[DATA_EXPOSURE]: The diagnostic process requires reading codebase files, logs, and artifacts (like HAR files) to understand failure modes. This access is inherent to the debugging task and does not show signs of sensitive data harvesting or network exfiltration.
[PROMPT_INJECTION]: The skill handles an indirect prompt injection surface by processing external data such as user bug reports and project code.
Ingestion points: User descriptions of bugs, codebase files, and tool outputs (stdout/stderr) are incorporated into the diagnostic context.
Boundary markers: The instructions do not define specific delimiters for separating untrusted data from agent instructions.
Capability inventory: The skill has the capability to execute shell commands (Phase 1), modify the codebase (Phase 5), and perform git operations.
Sanitization: There is no explicit sanitization of input data; however, the methodology requires a verifiable 'feedback loop' (Phase 1) and 'falsifiable hypotheses' (Phase 3), which serve as functional validation steps that mitigate the impact of potentially misleading data.

rightcode-diagnose