The Agent Skills Directory

[METADATA_POISONING]: The skill metadata in the YAML frontmatter contains misleading information. Specifically, the createdBy field identifies a non-existent model version ('gpt-5.6-sol') and the createdAt/updatedAt fields refer to dates in the year 2026. This inconsistency is a form of metadata deception.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from the user's local project environment, which creates a potential surface for indirect prompt injection.
Ingestion points: The scripts/inspect-project.py script reads text from build.gradle, settings.gradle, and libs.versions.toml files found in the project root.
Boundary markers: The script's output is structured as JSON, but the skill instructions do not provide explicit delimiters or instructions for the agent to ignore potential commands embedded within the analyzed project files.
Capability inventory: The skill allows the execution of a provided Python script and instructs the agent to run Gradle wrapper commands (./gradlew) on the host system.
Sanitization: The inspection script utilizes regular expressions to filter for specific SKIE and Kotlin configuration patterns and uses json.dumps to ensure the resulting data is formatted correctly as JSON.

skie