SKILL: Autocompact Circuit Breaker

Domain: context-management Trigger: Apply when autocompact failures repeat and the service must stop retrying to avoid API storms. Source Pattern: Distilled from reviewed session memory, compaction, and context-budgeting implementations.

Core Method

Track consecutive autocompact failures via the caller-provided auto compact tracking state. After each failure, increment the counter, log a warning once the configured limit is reached, and return the updated count instead of throwing. The loop uses this counter to skip further auto-compact attempts as long as the limit is exceeded, effectively acting as a circuit breaker that keeps doomed sessions from flooding the service.

Key Rules

• Add one to the failure counter on every exception (unless it’s a user abort).
• When the counter reaches the policy limit, log the breaker trip so telemetry reveals the gap.
• Return was compacted false plus consecutive failures so callers can persist the new state.
• Use the counter to skip auto-compact early on future turns rather than re-running the failed path.
• Reset the counter on a successful compaction so normal behavior resumes.

Example Application

After three consecutive failures hitting the context limit, auto compact if needed stops retrying, logs circuit breaker tripped, and the tracker causes the next turn to skip compaction until a successful run resets the counter.

Anti-Patterns (What NOT to do)

• Don’t let the controller keep retrying with the same broken context; the breaker exists to keep it quiet.
• Don’t swallow failure logs — future debugging relies on the warning emitted when the limit trips.