The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's README recommends an installation process that fetches the SKILL.md configuration from the author's official GitHub repository, which is a common and accepted practice for agent skills.
[COMMAND_EXECUTION]: To achieve its debugging goals, the skill executes various shell commands for version control (git), process isolation (nohup), and the execution of dynamically created reproduction scripts (e.g., repro.js).
[DATA_EXFILTRATION]: The skill performs local project discovery and log analysis, which involves reading configuration files and runtime output. These activities are contained within the local environment and are necessary for identifying the root causes of software issues.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it processes untrusted data from the project's source code and logs during the analysis phase. Ingestion points: Phase 1 (Discovery) and Phase 5 (Analysis). Boundary markers: None explicitly enforced beyond the use of tagged logging formats. Capability inventory: The agent can execute shell commands, manage background processes, and create/modify local files. Sanitization: The skill mitigates context window overflow and accidental instruction obedience by using grep, head, and tail to selectively extract relevant log data.

debug-mode