@bohuyeshan/openagent-labforge-core 3.13.1 → 3.13.2

package/README.md

@@ -1,235 +1,235 @@
-# OpenAgent Labforge
-
-OpenAgent Labforge is an OpenCode plugin fork focused on three things:
-
-- stronger engineering-oriented orchestration
-- explicit, inspectable child-session delegation
-- bioinformatics-first specialist workflows
-
-It is a derivative of `code-yeongyu/oh-my-openagent` and keeps upstream
-license/provenance boundaries intact. See [LICENSE.md](LICENSE.md),
-[NOTICE](NOTICE), [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md), and
-[docs/licensing.md](docs/licensing.md).
-
-## What This Fork Is Now
-
-This README describes current runtime behavior only.
-
-The plugin is now centered on:
-
-- OpenCode-native delegation through `task(subagent_type=...)`
-- stable plugin/agent/MCP injection for newer OpenCode behavior
-- explicit search/doc/code/paper retrieval separation
-- a growing first-party bioinformatics agent and skill stack
-- local-first development and installation
-
-## Current Core Capabilities
-
-### Engineering orchestration
-
-- `sisyphus`: main orchestrator
-- `wase`: fully autonomous orchestrator
-- `hephaestus`: deep coding executor
-- `prometheus`: planner
-- `atlas`: execution coordinator
-- `metis`: pre-planning consultant
-- `momus`: plan reviewer
-
-These core agents are being upgraded with stronger engineering behavior:
-
-- better scope discipline
-- stronger verification expectations
-- more explicit planning and review standards
-- cleaner delegation contracts
-
-Current engineering-capability layering:
-
-- strong execution + orchestration: `sisyphus`, `wase`
-- strong execution: `hephaestus`
-- strong orchestration: `atlas`
-- strong planning: `prometheus`, `metis`
-- strong review: `momus`
-
-This layering is intentional so future de-duplication stays easy if OpenCode
-itself absorbs parts of the same capability set.
-
-### Specialist agents
-
-- `explore`: local codebase discovery
-- `librarian`: focused upstream docs / SDK / library research
-- `github-scout`: repository and implementation scouting
-- `tech-scout`: ecosystem / benchmark / launch analysis
-- `article-writer`: public technical writing
-- `scientific-writer`: research-facing technical writing
-- `multimodal-looker`: PDF / image / figure understanding
-
-### Bioinformatics stack
-
-The fork now has an explicit bioinformatics hierarchy:
-
-- main bio entrypoints:
-  `bio-orchestrator` for integrated coordination
-  `bio-pipeline-operator` for execution-focused work
-- internal specialists used through child-session delegation:
-  `bio-methodologist` for computational design / QC / statistics planning
-  `wet-lab-designer` for user-executed wet-lab validation planning
-  `paper-evidence-synthesizer` for cross-paper evidence matrix and confidence grading
-
-This is not limited to generic "analysis". The intended workflow includes:
-
-- literature retrieval
-- public dataset discovery
-- computational analysis
-- wet-lab validation design for the user to execute
-- evidence integration
-- specialist reporting
-
-## Built-in Skill Direction
-
-The built-in skill set now covers both general and bio workflows.
-
-Examples:
-
-- browser / UI / git:
-  - `playwright`
-  - `frontend-ui-ux`
-  - `git-master`
-- document / report:
-  - `docx-workbench`
-  - `pdf-toolkit`
-  - `xlsx-analyst`
-- bioinformatics:
-  - `bio-tools`
-  - `bio-methods`
-  - `wet-lab-design`
-  - `bio-pipeline`
-- `paper-evidence`
-- `differential-expression`
-- `scrna-preprocessing`
-- `cell-annotation`
-- `pubmed-search`
-- `geo-query`
-- `sequence-analysis`
-- `structural-biology`
-- `bio-visualization`
-- `vector-design`
-
-The bio skills are written as execution-oriented references: they specify
-preferred tools, typical commands/code paths, expected artifacts, and boundary
-conditions rather than acting as vague prompt decoration.
-
-Bio agents now also carry explicit data-interaction and environment-safety behavior:
-
-- they ask users for the minimum decisive data when company or sequencing data is missing
-- they separate required inputs from optional context
-- they prefer `uv` for Python environments
-- they prefer `conda` for mixed native stacks
-- they call out when Windows users realistically need WSL/Linux
-
-## Current MCP Set
-
-The built-in MCP surface is intentionally tighter now.
-
-Default built-ins kept visible:
-
-- `websearch`
-- `context7`
-- `grep_app`
-- `browser_puppeteer`
-- `chrome-devtools-mcp`
-- `deepwiki_mcp` (default off)
-- `open_websearch_mcp`
-- `paper_search_mcp`
-- `semantic_scholar_fastmcp`
-
-Removed from the built-in visible set:
-
-- `arxiv_mcp`
-- `fetch_browser`
-
-Rationale:
-
-- avoid duplicate capability surfaces
-- keep the MCP list focused
-- leave `deepwiki_mcp` opt-in instead of always-on
-
-## OpenCode Compatibility Work
-
-Recent migration work has focused on keeping pace with newer OpenCode behavior.
-
-Already aligned or hardened:
-
-- runtime plugin config discovery with canonical + legacy config compatibility
-- agent registration and protected builtin-agent override handling
-- command discovery:
-  - `.opencode/command`
-  - `.opencode/commands`
-  - ancestor discovery up to worktree root
-  - slash-style nested command names
-- `.agents/skills` participation in injection chains
-- MCP merge order and user override behavior
-- todo continuation / compaction / stagnation guard regressions
-
-The migration audit reference lives at:
-
-- [docs/release/upstream-oh-my-openagent-3.11-plus-audit.md](docs/release/upstream-oh-my-openagent-3.11-plus-audit.md)
-
-## Installation Reality
-
-This project is still optimized for local-first use, but there is now a real
-published path for `Windows x64`.
-
-Current install reality:
-
-- `Windows x64`: published npm package path is available
-- other platforms: local build + local install remains the reliable path
-
-Recommended workflow:
-
-```bash
-bun run build:skills-catalog
-bun run build
-bun pm pack
-```
-
-Then follow:
-
-- [docs/guide/installation.md](docs/guide/installation.md)
-
-## Reference Repos
-
-Local comparison material is kept under `Future/` for migration and design work.
-
-Current references include:
-
-- upstream `oh-my-openagent`
-- `BioClaw`
-- `Geneclaw`
-- `codex-main`
-
-These are local reference trees, not part of the shipped plugin.
-
-## Current Priorities
-
-The near-term order is:
-
-1. finish upstream OpenCode compatibility migration
-2. strengthen engineering execution/orchestration capability in core agents
-3. continue refining the bioinformatics agent + skill stack
-
-## Docs
-
-- [docs/guide/installation.md](docs/guide/installation.md)
-- [docs/guide/orchestration.md](docs/guide/orchestration.md)
-- [docs/guide/bio-skills.md](docs/guide/bio-skills.md)
-- [docs/guide/bio-paper-autonomous-flow-v1.md](docs/guide/bio-paper-autonomous-flow-v1.md)
-- [docs/reference/configuration.md](docs/reference/configuration.md)
-- [docs/reference/features.md](docs/reference/features.md)
-
-## Language Versions
-
-- [English](README.md)
-- [简体中文](README.zh-cn.md)
-- [日本語](README.ja.md)
-- [한국어](README.ko.md)
-- [Русский](README.ru.md)
+# OpenAgent Labforge
+
+OpenAgent Labforge is an OpenCode plugin fork focused on three things:
+
+- stronger engineering-oriented orchestration
+- explicit, inspectable child-session delegation
+- bioinformatics-first specialist workflows
+
+It is a derivative of `code-yeongyu/oh-my-openagent` and keeps upstream
+license/provenance boundaries intact. See [LICENSE.md](LICENSE.md),
+[NOTICE](NOTICE), [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md), and
+[docs/licensing.md](docs/licensing.md).
+
+## What This Fork Is Now
+
+This README describes current runtime behavior only.
+
+The plugin is now centered on:
+
+- OpenCode-native delegation through `task(subagent_type=...)`
+- stable plugin/agent/MCP injection for newer OpenCode behavior
+- explicit search/doc/code/paper retrieval separation
+- a growing first-party bioinformatics agent and skill stack
+- local-first development and installation
+
+## Current Core Capabilities
+
+### Engineering orchestration
+
+- `sisyphus`: main orchestrator
+- `wase`: fully autonomous orchestrator
+- `hephaestus`: deep coding executor
+- `prometheus`: planner
+- `atlas`: execution coordinator
+- `metis`: pre-planning consultant
+- `momus`: plan reviewer
+
+These core agents are being upgraded with stronger engineering behavior:
+
+- better scope discipline
+- stronger verification expectations
+- more explicit planning and review standards
+- cleaner delegation contracts
+
+Current engineering-capability layering:
+
+- strong execution + orchestration: `sisyphus`, `wase`
+- strong execution: `hephaestus`
+- strong orchestration: `atlas`
+- strong planning: `prometheus`, `metis`
+- strong review: `momus`
+
+This layering is intentional so future de-duplication stays easy if OpenCode
+itself absorbs parts of the same capability set.
+
+### Specialist agents
+
+- `explore`: local codebase discovery
+- `librarian`: focused upstream docs / SDK / library research
+- `github-scout`: repository and implementation scouting
+- `tech-scout`: ecosystem / benchmark / launch analysis
+- `article-writer`: public technical writing
+- `scientific-writer`: research-facing technical writing
+- `multimodal-looker`: PDF / image / figure understanding
+
+### Bioinformatics stack
+
+The fork now has an explicit bioinformatics hierarchy:
+
+- main bio entrypoints:
+  `bio-orchestrator` for integrated coordination
+  `bio-pipeline-operator` for execution-focused work
+- internal specialists used through child-session delegation:
+  `bio-methodologist` for computational design / QC / statistics planning
+  `wet-lab-designer` for user-executed wet-lab validation planning
+  `paper-evidence-synthesizer` for cross-paper evidence matrix and confidence grading
+
+This is not limited to generic "analysis". The intended workflow includes:
+
+- literature retrieval
+- public dataset discovery
+- computational analysis
+- wet-lab validation design for the user to execute
+- evidence integration
+- specialist reporting
+
+## Built-in Skill Direction
+
+The built-in skill set now covers both general and bio workflows.
+
+Examples:
+
+- browser / UI / git:
+  - `playwright`
+  - `frontend-ui-ux`
+  - `git-master`
+- document / report:
+  - `docx-workbench`
+  - `pdf-toolkit`
+  - `xlsx-analyst`
+- bioinformatics:
+  - `bio-tools`
+  - `bio-methods`
+  - `wet-lab-design`
+  - `bio-pipeline`
+- `paper-evidence`
+- `differential-expression`
+- `scrna-preprocessing`
+- `cell-annotation`
+- `pubmed-search`
+- `geo-query`
+- `sequence-analysis`
+- `structural-biology`
+- `bio-visualization`
+- `vector-design`
+
+The bio skills are written as execution-oriented references: they specify
+preferred tools, typical commands/code paths, expected artifacts, and boundary
+conditions rather than acting as vague prompt decoration.
+
+Bio agents now also carry explicit data-interaction and environment-safety behavior:
+
+- they ask users for the minimum decisive data when company or sequencing data is missing
+- they separate required inputs from optional context
+- they prefer `uv` for Python environments
+- they prefer `conda` for mixed native stacks
+- they call out when Windows users realistically need WSL/Linux
+
+## Current MCP Set
+
+The built-in MCP surface is intentionally tighter now.
+
+Default built-ins kept visible:
+
+- `websearch`
+- `context7`
+- `grep_app`
+- `browser_puppeteer`
+- `chrome-devtools-mcp`
+- `deepwiki_mcp` (default off)
+- `open_websearch_mcp`
+- `paper_search_mcp`
+- `semantic_scholar_fastmcp`
+
+Removed from the built-in visible set:
+
+- `arxiv_mcp`
+- `fetch_browser`
+
+Rationale:
+
+- avoid duplicate capability surfaces
+- keep the MCP list focused
+- leave `deepwiki_mcp` opt-in instead of always-on
+
+## OpenCode Compatibility Work
+
+Recent migration work has focused on keeping pace with newer OpenCode behavior.
+
+Already aligned or hardened:
+
+- runtime plugin config discovery with canonical + legacy config compatibility
+- agent registration and protected builtin-agent override handling
+- command discovery:
+  - `.opencode/command`
+  - `.opencode/commands`
+  - ancestor discovery up to worktree root
+  - slash-style nested command names
+- `.agents/skills` participation in injection chains
+- MCP merge order and user override behavior
+- todo continuation / compaction / stagnation guard regressions
+
+The migration audit reference lives at:
+
+- [docs/release/upstream-oh-my-openagent-3.11-plus-audit.md](docs/release/upstream-oh-my-openagent-3.11-plus-audit.md)
+
+## Installation Reality
+
+This project is still optimized for local-first use, but there is now a real
+published path for `Windows x64`.
+
+Current install reality:
+
+- `Windows x64`: published npm package path is available
+- other platforms: local build + local install remains the reliable path
+
+Recommended workflow:
+
+```bash
+bun run build:skills-catalog
+bun run build
+bun pm pack
+```
+
+Then follow:
+
+- [docs/guide/installation.md](docs/guide/installation.md)
+
+## Reference Repos
+
+Local comparison material is kept under `Future/` for migration and design work.
+
+Current references include:
+
+- upstream `oh-my-openagent`
+- `BioClaw`
+- `Geneclaw`
+- `codex-main`
+
+These are local reference trees, not part of the shipped plugin.
+
+## Current Priorities
+
+The near-term order is:
+
+1. finish upstream OpenCode compatibility migration
+2. strengthen engineering execution/orchestration capability in core agents
+3. continue refining the bioinformatics agent + skill stack
+
+## Docs
+
+- [docs/guide/installation.md](docs/guide/installation.md)
+- [docs/guide/orchestration.md](docs/guide/orchestration.md)
+- [docs/guide/bio-skills.md](docs/guide/bio-skills.md)
+- [docs/guide/bio-paper-autonomous-flow-v1.md](docs/guide/bio-paper-autonomous-flow-v1.md)
+- [docs/reference/configuration.md](docs/reference/configuration.md)
+- [docs/reference/features.md](docs/reference/features.md)
+
+## Language Versions
+
+- [English](README.md)
+- [简体中文](README.zh-cn.md)
+- [日本語](README.ja.md)
+- [한국어](README.ko.md)
+- [Русский](README.ru.md)