What problem does it solve? Insights about how AI skills could be improved get lost between sessions because nobody captures them during real work. This Skill provides a persistent observation layer that watches multi-step tasks, user corrections, and methodology discussions, then logs reusable patterns to a persistent observation log so they can later be turned into new or improved skills. ## Core Features & Use Cases - Continuous Observation Logging: Silently appends structured observations (Issue, Suggested improvement, Principle) to a persistent log file throughout the entire session, including post-task feedback phases. - Collision-Safe Numbering: Enforces pre-write assertion and post-write verification with automatic renumbering to prevent duplicate observation IDs when parallel sessions write to the same log. - Skill Taxonomy Guidance: Classifies observations as open-source or internal skill candidates, with rules for attribution, licensing, lean content, and confidentiality boundaries. - Use Case: During a long agentic workflow, a user corrects the agent's output twice in the same way. The Skill logs the correction pattern immediately, and at the weekly review the pattern becomes a new enforcement rule in an existing skill. ## Quick Start Invoke the task-observer skill at the start of any session where you will use tools and produce deliverables, and add its activation instruction to your CLAUDE.md file.
