The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from JSON reports to generate instructions that the agent is expected to execute. If these reports originate from untrusted sources, they could contain malicious instructions designed to manipulate agent behavior.
Ingestion points: The scripts/fit_repair_queue.py script reads external content from files passed via the --fit-report and --texture-lighting-report arguments.
Boundary markers: Absent. The skill does not implement delimiters or specific instructions to the agent to treat the ingested data as untrusted content.
Capability inventory: The agent has access to highly capable tools including Bash and mcp__blender__execute_blender_code, which allow for shell command execution and arbitrary Python execution within the Blender environment.
Sanitization: Absent. The Python script performs no sanitization or validation on the keys or values extracted from the JSON reports before interpolating them into the output plan files.
[DYNAMIC_EXECUTION]: The SKILL.md file defines a Skill-gap rule which explicitly directs the agent to "patch the relevant skill(s)" if a failure repeats twice. This instruction encourages self-modifying behavior, where the agent alters its own instruction set or logic files at runtime, potentially leading to unintended or unsafe states.

fit-repair-optimizer