What problem does it solve? Long-running AI agents lose context between sessions, causing repeated investigations and forgotten decisions. This Skill standardizes how the Gemini CLI Bot persists state across sessions using a structured lessons-learned.md file. ## Core Features & Use Cases - Task Ledger Synchronization: Reads and updates a rolling ledger of the 50 most recent tasks, verifying PR and issue states via the GitHub CLI before starting work. - Decision & Hypothesis Logging: Records root-cause analyses, hypotheses, and architectural decisions in append-only logs with pruning rules to prevent context bloat. - Sub-agent State Delegation: Defines rules for passing ledger context to worker agents while restricting them to read-only memory access. - Use Case: A scheduled bot run begins by reading lessons-learned.md, marks merged PRs as DONE, investigates a failed task, then appends its findings and decisions before opening a new PR. ## Quick Start Activate the memory skill to synchronize the bot's task ledger with lessons-learned.md before starting the investigation and record all findings when finished.
