The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructs the agent to use sudo for high-risk system modifications, such as adding root certificates to the macOS system keychain (sudo security add-trusted-cert) and disabling Address Space Layout Randomization (ASLR) on Linux (echo 0 | sudo tee /proc/sys/kernel/randomize_va_space). These actions significantly weaken the host system's security posture.\n- [COMMAND_EXECUTION]: The methodology relies on broad shell command execution, including potentially destructive operations like pkill, rm, and dd for binary patching. For instance, it provides a command to overwrite binary data: printf '\\x90\\x90' | dd of=./target bs=1 seek=$((0x1234)) conv=notrunc.\n- [REMOTE_CODE_EXECUTION]: The agent is instructed to download and install a wide variety of tools and libraries via pip, npm, brew, go install, and gem. This introduces supply chain risks through the execution of remote code during the installation process.\n- [DYNAMIC_EXECUTION]: The skill includes custom scripts like dap.mjs that dynamically spawn processes using child_process.spawn. It also provides Python snippets to be executed via python3 -c for tasks like binary parsing and extraction.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes extensive external data (application logs, debugger outputs, network traces, and DOM snapshots) to drive its decision-making loop. This creates a vulnerability where malicious content in the analyzed data could trick the agent into executing unintended commands. * Ingestion points: references/methodology/02-investigate.md (logs), references/runtimes/native-binary.md (symbols), references/tools/browser-qa.md (DOM). * Boundary markers: None implemented for these data ingestion points. * Capability inventory: Shell access, subprocess spawning, file system modification, and network requests. * Sanitization: No sanitization or validation of the processed data is specified.\n- [EXTERNAL_DOWNLOADS]: The skill directs the agent to fetch resources from various external platforms, including GitHub, npm, and specialized tool domains such as frida.re and mitmproxy.org.\n- [PERSISTENCE]: The skill suggests hiding its debug journal by adding it to .git/info/exclude, which suppresses visibility of the artifact in standard git commands without modifying committed project files.

debugging