The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection when processing external code files.
Ingestion points: The audit, document, and redesign commands are designed to ingest and analyze source code from files specified by the user.
Boundary markers: Absent. The instructions do not define delimiters or explicit 'ignore instructions' markers to isolate untrusted data from the agent's core logic.
Capability inventory: The skill can execute local scripts via Node.js, perform shell commands, and write files to the filesystem.
Sanitization: Absent. There are no instructions to sanitize or escape content extracted from target files before processing.
[DYNAMIC_EXECUTION]: The skill uses runtime execution of local tools and shell commands to provide diagnostics.
Runtime script execution: The audit command instructs the agent to execute a bundled Node.js utility (scripts/detect.mjs) to scan files for design slop.
Parameter injection: The execution command node "$DETECT" --json <target ...> directly incorporates user-provided file paths into a shell execution string without validation instructions.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands to support its features.
Seed generation: The divergence logic requires the agent to run the shell command date +%S to generate a random seed for style rotation.
Tool discovery: The audit feature uses a shell loop to resolve the local path of the diagnostic script before execution.

finesse-ui