Note: telemetry

Context

The attune.telemetry package covers four distinct concerns that share a common infrastructure layer:

• Usage tracking — records help queries to help_queries.jsonl (log version 1.0) and exposes cost and cache statistics via the CLI
• Agent coordination — routes TTL-scoped signals between agents through CoordinationSignals (backed by Redis)
• Heartbeat tracking — lets HeartbeatCoordinator detect stale agents using Redis TTL keys
• Human approval gates — suspends workflow execution until a human responds via ApprovalGate

The feedback loop described in concepts/feedback-loop.md (FeedbackLoop, FeedbackEntry, QualityStats) is part of this package and is exported alongside the coordination and tracking classes.

Design

The package exposes classes and CLI entry points at the same boundary. The CLI functions in cli_analysis.py and cli_automation.py (for example, cmd_sonnet_opus_analysis, cmd_tier1_status) operate on the same data structures that the classes produce — AgentHeartbeat, CoordinationSignal, ApprovalRequest, and their counterparts. Neither layer wraps the other; they share the underlying Redis-backed memory store passed at construction time.

CoordinationSignal carries a ttl_seconds field (default 60) that controls how long a signal remains visible to wait_for_signal and check_signal. Signals past their TTL are invisible to consumers but are not automatically deleted; clear_signals() must be called explicitly to remove them.

Source

src/attune/telemetry/** — 16 source files total.

Tags: telemetry, metrics