The Agent Skills Directory

[SAFE]: The skill implements a security-first debugging workflow that mandates root-cause investigation before any code changes and explicitly treats all external data, such as logs and tool output, as untrusted content.\n- [SAFE]: Includes a 'careful' module that acts as a guardrail by intercepting and warning against potentially destructive shell commands such as recursive deletions, database drops, and forced git pushes.\n- [SAFE]: Provides a 'freeze' mechanism that restricts the agent's editing capabilities to a user-specified directory, preventing accidental modifications to the broader filesystem or repository.\n- [EXTERNAL_DOWNLOADS]: The runtime-bootstrap.mjs script downloads runtime components from the official 'time-attack/gstack' GitHub repository. These downloads are verified using SHA-256 hashes and Cosign attestations, ensuring the integrity of the code before installation.\n- [DATA_EXFILTRATION]: While the skill can perform web searches and use 'Context.dev' for research, it mandates the removal of sensitive identifiers like IPs, file paths, and customer data before transmission and requires per-session user consent.\n- [COMMAND_EXECUTION]: The skill uses standard system utilities such as git, grep, and xcodebuild for its core debugging and build tasks, which is consistent with its intended diagnostic and repair functions.

debug