The Agent Skills Directory

[COMMAND_EXECUTION]: The dbg launch command documented in SKILL.md and references/commands.md allows the agent to execute arbitrary shell commands to start debugging sessions. This capability provides a direct path for the agent to run any executable on the host system.
[REMOTE_CODE_EXECUTION]: Through the dbg eval and dbg hotpatch commands (detailed in references/commands.md), the agent can execute arbitrary code within a running process or modify its source code at runtime. These features represent significant execution capabilities within the target environment.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection (Category 8) as it lacks safeguards when processing untrusted runtime data.
Ingestion points: The agent retrieves data from external processes via dbg state, dbg vars, dbg props, dbg eval, and dbg console (found in references/commands.md).
Boundary markers: The documentation does not provide delimiters or warnings to treat debugged data as untrusted, which may lead the agent to interpret data as instructions.
Capability inventory: The skill enables high-impact actions such as dbg launch (subprocess spawning), dbg eval (dynamic code execution), and dbg hotpatch (file-system modification), as listed in SKILL.md.
Sanitization: There is no evidence of output sanitization or filtering to prevent malicious payloads in the debugged data from influencing the agent's behavior.

dbg