The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary workflow involves building and executing a shell command that incorporates multiple user-controlled or project-defined variables.
Evidence: In Step 3 of SKILL.md, the skill executes node "$OD_BIN" media generate using variables such as --model, --voice, and --prompt which are sourced from external metadata or user input.
Risk: There is no evidence of sanitization or parameterization of these inputs. An attacker providing a malicious prompt or metadata value (e.g., containing ;, $(...), or `) could achieve arbitrary command execution on the host system.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through project metadata.
Ingestion points: The skill reads audioKind, audioModel, audioDuration, and voice from project metadata in Step 0.
Boundary markers: No boundary markers or instructions to ignore embedded commands are present in the processing logic.
Capability inventory: The skill has access to a subprocess execution tool (node command).
Sanitization: The instructions explicitly tell the agent to use values from metadata "verbatim," which bypasses safety validation and directly introduces untrusted content into the execution pipeline.

audio-jingle