Releases
View all releases on GitHub →
v0.2.1 — April 1, 2026
Install
pipx install opentraces==0.2.1
# or
brew install JayFarei/opentraces/opentraces
Changes
- fix:
web_server.py now uses dynamic SCHEMA_VERSION instead of hardcoded "0.1.0" — web review UI was showing wrong schema version in trace previews
- fix: bundled
skill/SKILL.md now includes exit code 1 (needs review / quality gate not passed) for import-hf partial failures
- docs: post-release docs-update pass — viewer TypeScript types updated for v0.2.0 schema fields,
llms.txt regenerated from source, generate-llms-txt.sh updated with new workflow pages
- ci:
update-homebrew.yml now has workflow_dispatch for manual re-triggers; authenticated remote URL fix for push step
v0.2.0 — April 1, 2026
Install
pipx install opentraces==0.2.0
# or
brew install JayFarei/opentraces/opentraces
What's new
Runtime agent import
import-hf — pull traces from any HuggingFace dataset, parse via Hermes or generic adapter, and push to your own dataset
- Hermes parser — first-class support for Hermes-format agent logs
- Claude Code hook integration —
on_stop and on_compact hooks enrich sessions with context window state, project metadata, and tool usage at session end
Assessment workflow
push --assess — run quality scoring immediately after push
assess --dataset — re-assess a remote HF dataset without hf-mount; fetches JSONL directly via datasets-server API
- Scorecard badges — SVG badges for quality dimensions (training, RL, analytics, domain, conformance)
opentraces:stats sentinel — machine-readable stats block in dataset README; parsed by explore for community aggregation
CLI improvements
upgrade command — update opentraces in place
- Agent-aware exit codes: 0 = success, 1 = needs review, 2 = blocked by gate
--machine flag for structured JSON output consumable by agents
Schema 0.2.0
execution_context: "devtime" | "runtime" — discriminator for code-editing vs task-execution agents
Outcome.terminal_state — how the trajectory ended (goal_reached, interrupted, error, abandoned)
Outcome.reward and Outcome.reward_source — RL environment reward signals
- See opentraces-schema v0.2.0
Fixes
- Parser: harden
tool_input extraction against malformed model outputs
- Pipeline:
needs_review gates on scanner matches, not redaction count
- Upload: regenerate dataset card from full remote aggregate on each push
- HF: fix datasets-server indexing for multi-shard datasets
- Security: allowlist path-structured
high_entropy_string false positives
v0.1.1 — March 29, 2026
Install
pipx install opentraces
# or
brew install JayFarei/opentraces/opentraces
What's included
- Claude Code session parser with multi-agent adapter contract
- Context-aware security scanning and redaction
- Git signal enrichment and attribution
- Sharded JSONL upload to HuggingFace Hub
- Terminal UI (
lazytraces) and web review UI
- Per-project review policy (auto/review)
Changes since v0.1.0
- Fix publish workflow to correctly target TestPyPI vs PyPI
- Use
pipx for CLI install commands
- Add Homebrew tap auto-update on release