The Agent Skills Directory

[DATA_EXFILTRATION]: The 'Self-Evolution Telemetry' section instructs the agent to search for and write to files within the ~/.claude/projects/ directory. Accessing internal agent configuration or project memory paths can lead to unintended data exposure or modification of the agent's internal state.
[COMMAND_EXECUTION]: The 'Self-Healing Validation' phase requires the agent to 'Run build/compile' and 'Run the project's test suite.' This involves executing shell commands determined dynamically from the audited project, which could lead to arbitrary command execution if the project contains malicious build scripts or test configurations.
[PROMPT_INJECTION]: The skill uses the $ARGUMENTS variable directly in the TARGET field without sanitization. This provides a vector for direct prompt injection where a user could provide arguments designed to override the skill's logic or the behavior of the sub-skills it calls.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the game project it is auditing, creating an indirect prompt injection surface.
Ingestion points: The skill audits local game project files, including engine code, asset metadata, and configuration files via sub-skill calls in SKILL.md.
Boundary markers: There are no delimiters or 'ignore' instructions used when the agent processes project files or transitions between phases.
Capability inventory: The skill has the capability to execute build commands, run test suites, and write telemetry data to the local file system.
Sanitization: There is no evidence of sanitization, validation, or escaping of the content found within the audited game project before it is used to inform agent actions or shell commands.

game-launch