agent-traces

The living ecosystem where AI agents complete tasks through workflow loops, improve through iterative execution, are evaluated by mentor agents or humans in the loop, and turn completed work into reusable work experience and data to improve future agents.

Observability for AI agents. See what your agent did, why it cost that much, and what to fix.

Open skill for capturing AI agent work as structured traces.

Use cultivar to test your Agent Skills and Docs by running them in sandboxes, and across different agents.

Compile an AI agent's repeated workflows into deterministic, auditable routines that replay for free, with a fallback to the agent.

Generate realistic multi-agent workflow traces with LLM-enriched content, semantic validation, and PM4Py compatibility. pip install open-agent-traces

A governed learning layer for AI agents — turns execution traces into reviewed memories, reusable skills, and evidence-backed training data.

Agent serving trace from https://freeinference.org

📚 Curated catalog of agent-training datasets + a toolkit that normalizes, deduplicates, and quality-tiers them into one schema. Produces 🤗 voidful/agent-sft.

Agent Trace Analysis Platform (atap) — pluggable pipeline for trace representation, analysis, error classification, failure attribution and recovery for LLM agents

TraceCC gives AI agents a CLI-native long-term memory over their own session history. Searchable, scoped, and sized for context windows.

Accepted to NeurIPS 2026 (Main Track). Reviewer precision and critique uptake are separable: a multi-agent protocol can detect errors well and still fail to repair them. Code, data and analysis for arXiv:2607.15388.

Collect, normalize, audit, and render coding-agent traces in one portable format.

Unroll: read agent traces and chat logs in VS Code. Claude Code, Codex and other JSONL conversations as a readable chat. Public docs, examples and releases; implementation source is private.

Claude Code plugin that stops a coding agent from ending its turn with 'tests pass', 'build succeeds' or 'deployed' unless the transcript shows the command ran after the last edit and succeeded. Plus a CLI that audits agent transcripts for unsupported claims.

Retry Safety Analyzer: deterministic HIGH/MEDIUM/SAFE when a side-effecting tool may have been retried. Offline import — never executes the trace. HIGH is risk, not proof. External validation open.

Reference implementation, benchmarks, and traces for "Parsing the Stream: A Live Trace Model for Long-Horizon Agents and Their Observers" (arXiv:2609.01466). Append-only trace ledger → deterministic run state → per-consumer views for the human observer and the agent.

Codex plugin that turns long-running agent traces into personalized local audio briefings with OpenAI Realtime TTS

Governed agent runtime reference with policy gates, tool schemas, failure modes, and OpenTelemetry-shaped traces.

Stop reading agent logs. Watch them. Local viewer for Claude Code, Codex, Cursor, OpenCode, Grok and Droid sessions.