The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions direct the agent to search the codebase and evaluate performance hooks, which involves ingesting untrusted data from the repository.\n
Ingestion points: The agent performs recursive searches on the repository using the rg tool in SKILL.md and references/scenarios.md.\n
Boundary markers: There are no explicit instructions for the agent to distinguish between its internal instructions and the content retrieved from the repository.\n
Capability inventory: The skill environment allows for the execution of build commands (gradlew) and Python scripts (eval_triggers.py).\n
Sanitization: The skill does not define methods for validating or sanitizing the content discovered in the repository before the agent processes it.\n- [COMMAND_EXECUTION]: The skill documents commands to build projects and run evaluation scripts.\n
Evidence: SKILL.md and scripts/run_examples.sh contain instructions for executing ./gradlew :app:assembleDebug and python3 scripts/eval_triggers.py.\n- [EXTERNAL_DOWNLOADS]: The skill provides links to official developer documentation.\n
Evidence: SKILL.md and references/official-links.md link to developer.android.com for official Android performance and benchmarking guides.

android-performance-observability