v0.7.2

Added

• Experiment incident lessons: new IncidentLesson table, lessons MCP/service layer, and a project-level Lessons page. Lessons are surfaced to the experiment proposal flow so agents can avoid repeating known failure modes.
• Experiment report figure uploads: synapse_upload_document_image now accepts experiment-report figures and serves them from Synapse instead of relying on external image hosts.

Changed

• Release process: publishing the main npm package must now also rebuild and push the vincentwei1021/synapse Docker image (both :vX.Y.Z and :latest) so Docker deployments stay in sync with npm. Documented in CLAUDE.md and AGENTS.md.

v0.7.1 — CC plugin skills rewrite + onboarding alignment

Highlights

• CC plugin 0.7.0: full skill rewrite.
  • Sessions skill becomes a 10-hook mechanism manual with local state layout and parallel multi-experiment dispatch.
  • Autonomy skill becomes a Claude-Code-client loop — default full_auto, main agent orchestrates, sub-agents auto-spawn via the Task hook, mutual exclusion with the server-side autonomousLoopEnabled flag.
  • Experiments skill/reference gain foundational first-experiment guidance, a detailed execution checklist (tmux + unbuffered python + PEM + repo branch + cron), failure handling, and review-rejection flow.
  • Empty-project onboarding added across top-level, research, experiments, and autonomy skills.
  • Dropped agent-teams skill; content merged into sessions.
  • Removed OpenClaw-specific language from CC skills.
• Onboarding Step 2 (Claude Code): replaces the single claude mcp add command with a README-aligned three-block layout — plugin marketplace install + env vars + collapsed manual .mcp.json fallback. Matching en/zh i18n.
• Compute page: per-GPU Release button to force-clear stuck reservations.
• Documents: local-mode image upload; broken-image fallback; RQ → linked-experiment deep link.
• Settings: dropped Autonomous Loop and Auto Search cards; added synthesis reset control.
• Synthesis refresh: auto-completes on save; enriched full context.
• Experiments: clears liveStatus on manual exit from in_progress.
• Paper search: falls back to arXiv on 401 when DeepXiv token is absent.
• Reassign button: falls back to current assignee when the dropdown is untouched.

See CHANGELOG.md for the full list.

v0.7.0 — PGlite Local Mode

Highlights

One-command local install — no PostgreSQL or Redis required:

npm install -g @synapse-research/synapse
synapse

Changes

Bug Fixes

• Bug #19 (Critical): Research question creation silently failed in standalone mode — server actions replaced with REST API
• Bug #20 (Medium): Cache directory permission denied on global npm install — postinstall chmod fix
• Health check no longer false-reports "degraded" when Redis is configured but lazy-loaded

Improvements

• Completion notification toasts show a checkmark icon instead of a plain dot
• New /api/research-questions/[uuid]/review REST endpoint

Documentation

• README updated with one-command install instructions (EN + ZH)
• Consolidated E2E test report: 54 tests, all passing
• Removed test screenshots from repo root

CLI Package

• @synapse-research/synapse@0.2.5 on npm

v0.6.1

What's New

• Project descriptions on the dashboard now preserve intentional blank lines when expanded, improving readability for structured briefs
• The experiment plan side panel now renders above the detail sheet overlay, uses the shared panel background, and allows normal text selection
• The autonomous experiment entry button is more prominent, with refreshed blue styling and a yellow lightning icon for faster scanning
• The experiment detail sheet is wider on large screens, giving experiment plans, results, and progress logs more room

Notes

• This release also updates the English and Chinese README changelog sections for v0.6.1.

v0.6.0

Added

• Agent Type & Transport: Agents now have a type field (openclaw | claude_code) that maps to an internal transport capability (realtime | poll). Web UI dispatch features only show realtime-capable agents.
• Agent type selector: Create and edit forms on the Agents page include a Type dropdown with badge display.
• Transport query filter: GET /api/agents supports ?transport=realtime and ?type=claude_code query parameters.
• Dispatch validation: Auto-search and deep-research API routes reject agents that do not support realtime transport.
• Experiment assignments in checkin: synapse_checkin now returns assigned experiments alongside legacy experiment runs.
• Research Copilot workflow: Claude Code SessionStart hook presents research projects with progress summaries and a guided workflow covering paper search, deep research, research questions, experiments, and analysis.
• Project progress in checkin: synapse_checkin returns a projects array with paper counts, deep research status, research question titles, and experiment counts by status.

Changed

• Plugin version: Synapse Claude Code plugin bumped to 0.6.0.

Fixed

• Admin role validation: Added admin to valid agent roles in server actions and API routes.
• Type validation in server actions: createAgentAndKeyAction and updateAgentAction now validate agent type.
• Nullable auth guard: Fixed potential null auth context in integrations route.

v0.5.1

Added

• DeepXiv paper search: Replace Semantic Scholar + OpenAlex with DeepXiv hybrid search (BM25 + vector) over arXiv. arXiv API kept as fallback.
• Full-text paper reading: 4 new MCP tools for progressive reading — synapse_read_paper_brief (~500 tokens), synapse_read_paper_head (~1-2k), synapse_read_paper_section (~1-5k), synapse_read_paper_full (~10-50k)
• DeepXiv token in Settings: New Integrations card in Settings page — per-company DB-stored token with 1-min cache and env var override
• What's New in README: Both README.md and README.zh.md now have a What's New section updated each release

Changed

• Agent prompts: Auto-search uses brief for relevance assessment before adding papers; deep research uses progressive reading for full-text literature reviews
• OpenClaw plugin v0.9.0: New reading tool registrations and updated agent prompts

Removed

• Semantic Scholar & OpenAlex: Search adapters removed — paywalled papers can't be read anyway
• Stale screenshots: 14 debug screenshots (~2MB) removed from project root
• .vscode/: IDE config removed from repo, added to .gitignore
• CHANGELOG.md: Pre-rename Chorus-era changelog removed, replaced by README What's New

v0.5.0

Major Changes

Research & Literature

• Multi-source paper search (Semantic Scholar + OpenAlex + arXiv) with rate limiting, retry, and dedup
• Deep research report versioning (v1 → v2 → v3)
• Related Works page: one-shot paper search + deep research trigger with prompt editing
• Paper publishedYear extraction and display

MCP Tools & Permissions

• Tool cleanup: 114 → 65 tools (removed legacy ExperimentDesign/Run/Hypothesis)
• 5-permission model: pre_research, research, experiment, report, admin
• Tool names normalized (removed research_lead_ and pi_ prefixes)

GitHub Integration

• Project-level repo config (URL, username, token)
• Base branch selection when creating experiments
• Agent pushes code to experiment branches, tracked in UI
• synapse_get_repo_access MCP tool

Compute & GPU

• Per-node GPU telemetry toggle (30s polling, auto-disable after 3 failures)
• GPU reservation system (synapse_reserve_gpus) with atomic transactions
• Queued experiments automatically re-triggered on GPU release by priority

Experiment Workflow

• Structured experiment prompt: check GPU → reserve → start → SSH → clone → execute → monitor → push → submit
• Cron-based monitoring for long-running experiments
• liveStatus support in progress reports (queuing/checking_resources/running)

UI & Docs

• Dark mode color fixes across all pages
• Document types: experiment_result, literature_review, analysis, methodology
• GitHub config in project settings and edit dialog
• Deployment guide with OIDC/Cognito setup
• Three-environment sync rules in CLAUDE.md/AGENTS.md