Skip to content

v0.1.1

Choose a tag to compare

View all tags
@github-actions github-actions released this 26 Jul 04:58
· 147 commits to main since this release
v0.1.1
f06a65f

AgentOps Open Course v0.1.1

Patch release: publishes the "AgentOps Agent" rename and the pre-launch review fixes so the Latest release reflects current naming. Image digests and SPDX SBOMs for the agent and MLflow images are attached; images are cosign-signed (keyless, GitHub OIDC) and their signatures + SBOM attestations are verified in-workflow.

Changed

  • Renamed the reference agent from "Ops Copilot" to "AgentOps Agent", aligning the application identity (OTEL_SERVICE_NAME, MLflow experiment and prompt registry, ADK app_name, MCP server, audit actor, gateway backend) with the agentops-agent name the container image, Kubernetes workload, and Python distribution already used.
  • Promoted the default local model to qwen3:4b-instruct (Qwen3 4B Instruct 2507) for stronger tool calling at the same 2.5 GB footprint, and documented gemma4:e4b as an optional, heavier Apache-2.0 alternative.
  • Workflow sub-agents now enforce the same per-session token budget as every other agent (enforce_token_budget/record_token_usage).

Fixed

  • Corrected the front matter on the Overview, Quality, and Observability chapter indexes, where an unquoted colon made the YAML invalid and published the page description as a heading. scripts/check-docs.sh now parses front matter instead of pattern-matching it.
  • Made semantic runbook indexing atomic — a BEGIN IMMEDIATE rebuild — so two concurrent first-use turns can no longer race a parallel index drop/create.
  • Long-term memory now surfaces the same DataAccessError boundary as the primary data layer instead of leaking a raw SQLite driver error.
  • The prompt A/B evaluator reads a marked child-output line and re-raises the child's stderr on failure, instead of assuming its scores are the last line printed and swallowing the cause.
  • mise run config:check now names the offending field on a validation error, and eval:cost reports an actionable message for a malformed AGENT_COST_TOLERANCE rather than an uncaught ValueError.
  • Corrected the container build-stage count (three, not two), the ObservabilityCollectorDown runbook cross-reference, the Observability chapter description and incident loop, and several command-directory and cross-link notes across the course.
  • Aligned the agent-guardrails skill's kill-switch variable name to AGENT_WRITES_DISABLED, converted the agentops-course skill's cross-links to GitHub-rendering Markdown, and clarified the source-path roots in the installable skills.
Assets 6
Loading