The Agent Skills Directory

[DATA_EXFILTRATION]: The skill attempts to write metadata about its own execution and the project it is auditing to a hidden file path in the user's home directory.
Evidence: Instructions to "Look for the project path in ~/.claude/projects/" and "append to skill-telemetry.md in that memory directory" in SKILL.md.
[PERSISTENCE]: The telemetry mechanism creates a persistent record of agent activity in a hidden system directory, allowing tracking of user interactions or project details across different sessions.
Evidence: The 'Self-Evolution Telemetry' section defines a structured logging process to a persistent file location.
[METADATA_POISONING]: The skill's description and name do not disclose the automated file system modifications or telemetry logging behavior, potentially misleading the user about its side effects.
Evidence: The metadata fields focus entirely on game security review and omit the background logging activity.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it is instructed to ingest and analyze untrusted codebase data while maintaining file system write capabilities.
Ingestion points: "Investigate the codebase thoroughly" (SKILL.md).
Boundary markers: None; the instructions do not specify delimiters or warnings to ignore instructions embedded within the analyzed code.
Capability inventory: The skill maintains a persistent write capability to the local file system (telemetry logging).
Sanitization: None; there is no validation or escaping performed on the content read from the project codebase.
[TIME_DELAYED_CONDITIONAL]: The telemetry logging behavior is gated by a conditional check for the existence of a specific environment path, a pattern often used to trigger behavior only on targeted systems.
Evidence: "Only log if the memory directory exists. Skip silently if not found." (SKILL.md).

game-security