The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill explicitly checks for and reads ~/.claude/feishu.json. This file typically stores sensitive authentication tokens or webhook keys for Feishu notifications.
[COMMAND_EXECUTION]: The skill uses Bash(*) permissions to perform arbitrary file modifications, run experiments, and execute system commands during the 'Implement Fixes' phase.
[COMMAND_EXECUTION]: In 'Nightmare' mode, the skill executes a shell command using codex exec which passes a prompt to an external model specifically instructing it that it has 'FULL READ ACCESS' to the repository and that the local agent 'does NOT control what you see'.
[COMMAND_EXECUTION]: The workflow involves deploying code to GPU servers via SSH, which may involve the use of local SSH keys or credentials for remote access.
[EXTERNAL_DOWNLOADS]: The skill uses curl to fetch citation data from dblp.org and doi.org. While these are well-known academic services, they represent automated network activity to external domains.

auto-review-loop