ace-context-engineering

AI agents learn by reflecting on failures, proposing new guidance rules to prevent recurrence, and integrating them into their operational playbook. This Skill automates that evolution by retrieving relevant context bullets, applying incremental updates, and tracking confidence scores so agents self-correct without human intervention.

Progressive disclosure loads only the most relevant context when needed, preventing information overload. This Skill retrieves and prioritizes guidance bullets based on task relevance, reducing cognitive load and improving focus so agents execute faster and more accurately.

Provenance tracking with evidence maintains a record of which guidance bullets influenced each decision. This Skill validates bullets against schemas, scores their confidence, and logs the reasoning chain so you can audit agent behavior and identify sources of errors.

Yes. This Skill evolves task context automatically by retrieving relevant bullets each cycle, applying delta changes to refine guidance, and merging updates incrementally. It handles validation and confidence scoring so your agent's playbook improves without manual curation.

The Skill uses jsonschema to validate deltas and bullets against structured schemas, ensuring guidance updates are safe and consistent. It performs per-bullet confidence scoring and provenance tracking to maintain reliability as the playbook grows across iterations.

Use it for complex, iterative tasks where agents encounter new patterns repeatedly and benefit from learning. Static prompts work for simple, stable tasks; context evolution is best for long-running agents that must adapt to edge cases, refine strategies, and reduce repetitive failures autonomously.