The Agent Skills Directory

Dynamic Script Generation and Execution: The skill encourages the agent to develop and execute temporary local scripts to perform hypothesis testing and data slicing. This allows the agent to create new executable content at runtime based on its analysis of repository metrics.
Indirect Prompt Injection Surface: The agent processes repository metrics stored in CSV files, which may include data derived from external sources like issue titles or contributor comments. This creates a potential surface where untrusted data could influence the agent's analysis or the scripts it proposes.
Ingestion points: The skill reads data from tools/gemini-cli-bot/history/metrics-timeseries.csv and tools/gemini-cli-bot/history/metrics-before-prev.csv (SKILL.md).
Boundary markers: No specific delimiters or "ignore embedded instruction" warnings are mentioned for the metrics data being processed.
Capability inventory: The agent can use the gh CLI, GraphQL, generate and run local scripts, and propose changes to existing scripts in tools/gemini-cli-bot/metrics/scripts/ (SKILL.md).
Sanitization: The instructions do not specify sanitization or validation steps for the metrics data before it is used for hypothesis generation or script modification.
Self-Modifying Logic Considerations: The agent is authorized to analyze and propose changes to its own automation scripts located in tools/gemini-cli-bot/metrics/scripts/. This feedback loop, where analyzed data influences the agent's future logic, is a central part of the skill's repository health improvement mission but represents a capability that should be monitored for unexpected modifications.
Authorized Vendor Tooling: The skill references the use of bundle/gemini.js for classification tasks. This appears to be a tool provided by the vendor (google-gemini) as part of the bot's standard operating environment.

metrics