The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file references/install/java.md contains a hardcoded GitHub Personal Access Token. Hardcoding active credentials in documentation is a significant security risk.
[OBFUSCATION]: The GitHub Personal Access Token in references/install/java.md is split across two properties (syncause.repo.token.p1 and syncause.repo.token.p2), which is a technique typically used to bypass automated secret scanning tools.
[REMOTE_CODE_EXECUTION]: The Node.js installation process in references/install/nodejs.md involves piping a remote shell script directly into bash (curl | bash). This provides the script with full execution privileges on the user's machine.
[DYNAMIC_EXECUTION]: The Java installation (references/install/java.md) utilizes the byte-buddy-maven-plugin to perform bytecode transformations. This allows the SDK to inject logic into the application's classes at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves analyzing runtime traces and snapshots from external applications via MCP tools (e.g., get_trace_insight, inspect_method_snapshot). Since this data is generated by the application being debugged and could contain user-influenced logs or errors, it represents a surface for indirect prompt injection.
Ingestion points: Runtime trace data retrieved via get_trace_insight and inspect_method_snapshot (SKILL.md).
Boundary markers: None present; the skill treats all captured runtime state as trusted data.
Capability inventory: The skill has the ability to edit source code (Phase 2), execute shell commands for bug reproduction (Phase 1), and perform cleanup/uninstallation tasks (Phase 4).
Sanitization: No sanitization or filtering of the captured runtime data is performed before it is presented to the agent.
[EXTERNAL_DOWNLOADS]: The skill downloads and installs various artifacts from external sources, including a Python wheel from GitHub releases and multiple Node.js packages from the NPM registry.

syncause-debugger