Skip to content

Releases: lpc0387/hermes-collab-engine

Release list

v6.0 — leader=hermes, aggregate=hermes, 经验总结独立化

v6.0 — leader=hermes, aggregate=hermes, 经验总结独立化 Latest
Latest

Choose a tag to compare

View all tags
@lpc0387 lpc0387 released this 22 Jun 07:20
v6.0
77af50d

v6.0 正式版本

架构变更

  • Leader=Hermes, Worker=OpenCode — 双 Agent 分离，WBS/aggregate 用 hermes 闭环
  • Planner 重构 — WBS 分解由 hermes 完成，不再依赖 claude
  • Aggregate 用 hermes — 汇总报告也走 hermes，形成完整闭环

经验总结独立化

  • 每日 lessons 自动归档到 data/memory/
  • 回填前一天缺失的总结
  • 无 Hermes 路径依赖

其他

  • MCP 集成 puppeteer 浏览器自动化（7 个工具）
  • Lesson 质量体系（L1/L2/L3 + tags）
  • 安装脚本依赖更新（hermes + opencode 为必需）
Assets 2
Loading

Hermes Collab Engine v5.0.0 — First Formal Public Release

Hermes Collab Engine v5.0.0 — First Formal Public Release

Choose a tag to compare

View all tags
@lpc0387 lpc0387 released this 13 Jun 10:01
v5.0.0
eacd486

Hermes Collab Engine v5.0.0

Hermes Collab Engine v5.0.0 is the first formal public release of the standalone Hermes collaboration workflow. It packages the pre-release v4.5 capability set into a public baseline for review: WBS planning, parallel Worker execution, dashboard observability, sandbox demos, and local verification.

Highlights

  • Leader-led WBS collaboration — the Leader scores request complexity, decomposes work into dependency-aware WBS nodes, dispatches Workers, and aggregates node results into a final response.
  • Parallel Worker execution — Worker nodes can run concurrently within configured concurrency and timeout limits, using selected Skills and tool profiles for the node type.
  • Real-time dashboard — inspect runs, nodes, Worker pool state, selected Leader/Worker models, injected Skills/Tools, logs, and aggregate Leader feedback.
  • Leader feedback diary — completed runs can show a pixel notebook-style feedback view with copy and Markdown download actions.
  • Agent Backend abstraction — Claude Code, Codex, OpenCode, and custom command-backed coding agents can be registered behind one engine interface.
  • Install and integration templates — one-line install prepares a local checkout and virtual environment; Hermes integration templates are reviewed via dry-run before use.
  • Isolated sandbox demo — launch a mock-data dashboard demo with TTL cleanup, optional sub-path deployment, and optional limited real-worker execution in sandbox-only runtime paths.
  • Release safety boundary — the repository ships code, templates, and sanitized demo paths only; it does not include real runtime databases, secrets, tokens, sessions, logs, memories, or auth files.

Demo and project links

New contributor tasks

Eight good first issue / help wanted tasks are open for contributors:

  • #1 Document common sandbox startup failures
  • #2 Add a concise Agent Backend example
  • #3 Add dashboard screenshot refresh guidance
  • #4 Add CLI status JSON regression coverage
  • #5 Add sandbox TTL edge-case tests
  • #6 Polish dashboard empty-state copy
  • #7 Add issue-template label guidance
  • #8 Add README localization alignment checklist

Issue list: https://github.com/lpc0387/hermes-collab-engine/issues

Contact and support

Install

One-line local install

curl -fsSL https://raw.githubusercontent.com/lpc0387/hermes-collab-engine/main/scripts/install.sh | bash
cd ~/hermes-collab-engine

The installer checks local dependencies, clones or updates the repository, creates a virtual environment, and creates empty template directories. It does not copy real Hermes/Claude secrets or runtime databases into the repository.

Manual developer install

git clone https://github.com/lpc0387/hermes-collab-engine.git
cd hermes-collab-engine
python3 -m pip install -e .

Review Hermes integration templates

cd ~/hermes-collab-engine
./scripts/install-hermes-integration.sh --dry-run

Run the integration script without --dry-run only after reviewing the files it would create or update.

Demo instructions

Start the sandbox dashboard

./scripts/start_sandbox.sh

Open http://127.0.0.1:8876/. The default sandbox uses mock API responses and sanitized demo data; it does not call real Workers or write production data.

Useful sandbox options:

./scripts/start_sandbox.sh 0.5            # 30 minutes
./scripts/start_sandbox.sh --hours 4      # custom TTL
./scripts/start_sandbox.sh --port 8877    # custom port
./scripts/start_sandbox.sh --no-reseed    # reuse existing demo DB
./scripts/start_sandbox.sh --real         # limited real workers in isolated DB/workspace

Real-worker sandbox mode is intentionally isolated: it writes to data/sandbox_real.sqlite3 and data/sandbox_workspace/, not production data.

Run a direct collaboration task

hermes-collab run "Analyze the current project structure" --cwd . --json

Start the dashboard server manually

hermes-collab server --host 0.0.0.0 --port 8765 --cwd .

Verification checklist

Run these checks from the repository root before tagging, packaging, or publishing a public announcement:

python3 -m py_compile src/hermes_collab_engine/*.py sandbox/server.py scripts/seed_demo_data.py
PYTHONPATH=src python3 -m unittest discover -s tests -v
python3 -m hermes_collab_engine.cli verify-release
bash -n scripts/install.sh scripts/install-hermes-integration.sh scripts/start_sandbox.sh start.sh

Known limitations

  • verify-v45 remains available as a compatibility alias because the verification checks originated in the v4.5 pre-release line; public release checks should use verify-release.
  • The sandbox is for demos and release review, not production hosting. Keep it bound to localhost or a reviewed reverse proxy and use sanitized data only.
  • Real Worker execution depends on locally installed Agent Backends and credentials. The public repository intentionally provides templates instead of real secrets/configuration.
  • Long-running or high-concurrency real tasks should be reviewed for timeout, tool-profile, and workspace limits before public demonstration.

Launch messaging

Short announcement:

Hermes Collab Engine v5.0.0 is the first formal public release of the standalone Hermes collaboration workflow: a Leader decomposes complex work into WBS nodes, Workers execute in parallel, and a real-time dashboard makes the full run observable from plan to aggregate feedback.

Expanded announcement:

Today we are publishing Hermes Collab Engine v5.0.0, the first formal public baseline for the Hermes multi-agent collaboration engine. It includes WBS planning, dependency-aware Worker execution, Skill/Tool injection, a live dashboard, Leader feedback diary, one-line install path, and an isolated sandbox demo. The release is designed for public review without shipping runtime secrets or production data: local credentials stay local, the sandbox defaults to mock data, and optional real-worker demos run in isolated sandbox paths.

Release lineage note

Internal v4.5 documentation and the legacy verify-v45 alias describe the pre-release capability milestone that v5.0.0 formalizes for public release. Refer to that lineage only when explaining compatibility or verification naming; public release notes should lead with v5.0.0 and verify-release.

Loading