The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill involves the agent reading and interpreting compiler-generated report files, which creates a surface for ingesting external data into the context.
Ingestion points: The agent is instructed to read app_release-classes.txt, app_release-composables.txt, and app_release-module.json from the module's build directory.
Boundary markers: The skill relies on searching for specific compiler flags (e.g., restartable, skippable) rather than using explicit boundary delimiters or 'ignore' warnings for the ingested content.
Capability inventory: The agent uses its file-reading capability to perform diagnosis; it does not perform any writes, network operations, or subprocess executions based on the contents of these reports.
Sanitization: No sanitization is performed on the ingested text artifacts as they are expected to be machine-generated by the local build process.
[COMMAND_EXECUTION]: The skill guides the developer to run standard build commands locally to generate diagnostic data.
Evidence: Instructions to execute ./gradlew :app:assembleRelease -PcomposeCompilerReports=true in SKILL.md.
The commands are standard development tasks for the Android ecosystem and are scoped to the primary purpose of the skill.

diagnosing-compose-stability