@whatasoda/agent-tools 0.1.0 → 0.1.1

package/dist/scripts/codex-review.js

@@ -172,12 +172,7 @@ async function main() {
     const sessionPart = await buildSessionPart(parsed.sessionJsonlPath, tmpDir);
     const prompt = `${parsed.instruction}: ${reviewFile} (ref: ${refPath})${sessionPart}`;
     try {
-      const { output, exitCode } = await runCodex([
-        "exec",
-        "-m",
-        "gpt-5.4",
-        prompt
-      ]);
+      const { output, exitCode } = await runCodex(["exec", "-m", "gpt-5.4", prompt]);
       if (exitCode !== 0) {
         if (exitCode === 126 || exitCode === 127) {
           console.error(`⚠ codex レビューをスキップします（codex コマンドが見つからない、または実行権限がありません — exit code: ${exitCode}）`);

package/dist/skills/soda-brief/body.md

@@ -21,12 +21,10 @@ Identify the core topic from $ARGUMENTS. Determine:
 
 ### Step 2: Survey Investigation
 
-Launch a sub-agent (Task, subagent_type: Explore) to survey the relevant area.
-
 **Sub-agent prompt constraints**: Every sub-agent prompt MUST begin with:
 > You are a research-only agent. Do NOT use AskUserQuestion, EnterPlanMode, or any interactive/planning tools. Return your findings in the output format specified below.
 
-**Sub-agent output contract**: Every sub-agent prompt MUST end with:
+**Codebase investigation output contract**: Every codebase sub-agent prompt MUST end with:
 > Return findings in this exact format:
 > ### Files
 > - `path/to/file` — relevance to the topic
@@ -37,7 +35,42 @@ Launch a sub-agent (Task, subagent_type: Explore) to survey the relevant area.
 > ### Open Questions
 > - question — what remains unclear from this investigation alone
 
-The survey prompt should include the topic and ask for: relevant files, existing patterns, dependencies, and current state of the area.
+**External research output contract**: Every external research sub-agent prompt MUST end with:
+> Return findings in this exact format:
+> ### Official Documentation
+> - library/service name — key API, configuration, version-specific notes
+> ### Best Practices
+> - practice — source and context
+> ### Patterns & Examples
+> - pattern — description with code snippets if available
+> ### Caveats
+> - caveat — gotchas, known issues, version incompatibilities
+
+#### External Research Trigger
+
+Before launching sub-agents, evaluate whether the topic involves external technologies:
+
+- If the topic references **named external libraries, frameworks, or services** (e.g., "Drizzle", "NextAuth", "Stripe"), **technology selection or comparison** (e.g., "which ORM", "migrate from X to Y"), or **external API integration** (e.g., "Slack API", "GitHub webhook") → launch an external research sub-agent **in parallel** with the codebase survey.
+- If the topic is purely about modifying existing code with no new external dependencies → skip external research.
+
+**External research sub-agent prompt template**:
+> You are a research-only agent. Do NOT use AskUserQuestion, EnterPlanMode, or any interactive/planning tools. Return your findings in the output format specified below.
+>
+> ## Research Task
+> Investigate external documentation and resources for: [topic extracted from $ARGUMENTS]
+>
+> ## Research Strategy
+> 1. Use Context7 MCP (resolve-library-id → get-library-docs) for each identified library/framework
+> 2. Use WebSearch for broader context: best practices, migration guides, comparison articles, known issues
+> 3. Synthesize findings — prioritize official documentation over community content
+>
+> [External research output contract]
+
+#### Codebase Survey
+
+Launch a sub-agent (Task, subagent_type: Explore) to survey the relevant area. The survey prompt should include the topic and ask for: relevant files, existing patterns, dependencies, and current state of the area.
+
+If external research is triggered, launch both sub-agents in a single message (parallel execution).
 
 ### Step 3: Focused Investigation (optional)
 
@@ -53,6 +86,7 @@ Synthesize findings into a Discussion Briefing block:
 ## Discussion Briefing
 - **Topic**: what needs to be discussed
 - **Background**: relevant codebase findings and current state
+- **External Context** (include if external research was performed): key findings from official docs, best practices, and caveats
 - **Key Questions**: 2-4 questions that should guide the discussion (ordered by dependency)
 - **Constraints**: known technical or design constraints
 ```

package/dist/skills/soda-discuss/body.md

@@ -171,6 +171,18 @@ These are flexible guidance, not mandatory steps. Adapt to the conversation.
 
 - **Start with understanding**: Grasp what the user wants to explore. If `$ARGUMENTS` is vague, ask clarifying questions — but don't over-interrogate. One or two questions is usually enough to get started.
 - **Investigate with sub-agents**: Use sub-agents (Task, subagent_type: Explore) for codebase investigation. Apply the standard constraint block (above). Investigation informs the discussion but doesn't replace it.
+- **External research when needed**: When the discussion involves external libraries, technology comparison, or API integration, launch an external research sub-agent in parallel with any codebase investigation. The external research sub-agent should use Context7 MCP (resolve-library-id → get-library-docs) for official documentation and WebSearch for broader context (best practices, migration guides, known issues). Apply the standard constraint block. Use the external research output contract:
+  > Return findings in this exact format:
+  > ### Official Documentation
+  > - library/service name — key API, configuration, version-specific notes
+  > ### Best Practices
+  > - practice — source and context
+  > ### Patterns & Examples
+  > - pattern — description with code snippets if available
+  > ### Caveats
+  > - caveat — gotchas, known issues, version incompatibilities
+
+  Present external research findings alongside codebase findings before asking for decisions — consistent with the "データが先、判断が後" principle.
 - **Iterate naturally**: Some discussions need multiple investigation rounds; others converge quickly. Follow the conversation's natural rhythm.
 - **Record decisions immediately**: When the user approves a design decision, persist it to the DB via `sd decision create`. The Bash tool permission prompt serves as a natural approval checkpoint. Include `--repo-owner` and `--repo-name` when working in a git repository, and `--tag topic:<topic-slug>` for grouping.
 - **Capture memos as text during discussion**: When a notable idea or insight emerges, present it as text ("メモ：〇〇") without writing to the DB. Memos are batched and written during wrap-up.

package/dist/skills/soda-plan/body.md

@@ -14,7 +14,7 @@ If $ARGUMENTS is empty, ask the user what they want to implement before proceedi
        - Extract decisions as **mandatory constraints** — every applicable decision must be traceable to a plan step
        - Extract `rejected_alternatives` from each decision as **exclusion constraints** — approaches listed as rejected must not be re-proposed
      - If no decisions found: proceed normally (conversation-context-based input)
-   - **Sub-agent output contract**: Every sub-agent prompt MUST end with the following output format requirement:
+   - **Codebase investigation output contract**: Every codebase sub-agent prompt MUST end with the following output format requirement:
        > Return findings in this exact format:
        > ### Files
        > - `path/to/file` — relevance to the task
@@ -24,11 +24,36 @@ If $ARGUMENTS is empty, ask the user what they want to implement before proceedi
        > - dependency — how it affects the task
        > ### Open Questions
        > - question — what remains unclear from this investigation alone
-     - Launch a sub-agent (Task, subagent_type: Explore) to survey project structure, dependencies, and conventions relevant to the task.
-     - Summarize the agent's findings into a Common Context block.
-     - Based on findings, optionally launch 1-2 focused sub-agents in parallel. Each prompt must include the Common Context block (summarized, not raw output), the specific investigation question, and both the constraint block and output contract.
+   - **External research output contract**: Every external research sub-agent prompt MUST end with:
+       > Return findings in this exact format:
+       > ### Official Documentation
+       > - library/service name — key API, configuration, version-specific notes
+       > ### Best Practices
+       > - practice — source and context
+       > ### Patterns & Examples
+       > - pattern — description with code snippets if available
+       > ### Caveats
+       > - caveat — gotchas, known issues, version incompatibilities
+   - **External Research Trigger**: Before launching the survey sub-agent, evaluate whether the task involves external technologies:
+     - If the task references **named external libraries, frameworks, or services**, **technology selection or comparison**, or **external API integration** → launch an external research sub-agent **in parallel** with the codebase survey.
+     - If the task is purely about modifying existing code with no new external dependencies → skip external research.
+     - External research sub-agent prompt:
+       > You are a research-only agent. Do NOT use AskUserQuestion, EnterPlanMode, or any interactive/planning tools. Return your findings in the output format specified below.
+       >
+       > ## Research Task
+       > Investigate external documentation and resources for: [topic]
+       >
+       > ## Research Strategy
+       > 1. Use Context7 MCP (resolve-library-id → get-library-docs) for each identified library/framework
+       > 2. Use WebSearch for broader context: best practices, migration guides, comparison articles, known issues
+       > 3. Synthesize findings — prioritize official documentation over community content
+       >
+       > [External research output contract]
    - **Sub-agent prompt constraints**: Every sub-agent prompt (both survey and focused) MUST begin with the following constraint block:
      > You are a research-only agent. Do NOT use AskUserQuestion, EnterPlanMode, or any interactive/planning tools. Return your findings in the output format specified below.
+   - Launch a sub-agent (Task, subagent_type: Explore) to survey project structure, dependencies, and conventions relevant to the task. If external research is triggered, launch both sub-agents in a single message (parallel execution).
+   - Summarize findings from all agents into a Common Context block. If external research was performed, include a dedicated "External Context" section in the Common Context block.
+   - Based on findings, optionally launch 1-2 focused sub-agents in parallel. Each prompt must include the Common Context block (summarized, not raw output), the specific investigation question, and both the constraint block and output contract.
    - Summarize investigation results before proceeding.
    - If investigation reveals multiple fundamentally different approaches, use AskUserQuestion to let the user decide: "方針を調整して続行" / "ここで中断して方針を整理". Do not choose an approach autonomously.
 2. **Branch Strategy**: Use AskUserQuestion to determine branch strategy before planning:

package/package.json

@@ -1,14 +1,21 @@
 {
   "name": "@whatasoda/agent-tools",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "private": false,
   "type": "module",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/whatasoda/agent-extensions"
+  },
   "bin": {
     "sd": "dist/src/cli.js"
   },
   "scripts": {
     "build": "bun scripts/build.ts",
-    "test": "bun test"
+    "test": "bun test",
+    "typecheck": "tsc --noEmit",
+    "lint": "oxlint src/ scripts/",
+    "format:check": "oxfmt --check src/ scripts/"
   },
   "files": [
     "dist/"
@@ -21,9 +28,11 @@
     "ulid": "latest"
   },
   "devDependencies": {
-    "bun-types": "latest",
     "@types/react": "latest",
-    "typescript": "latest",
-    "ink-testing-library": "latest"
+    "bun-types": "latest",
+    "ink-testing-library": "latest",
+    "oxfmt": "^0.43.0",
+    "oxlint": "^1.58.0",
+    "typescript": "latest"
   }
 }