omnilearn-workflow 1.0.0 → 1.0.2

package/commands/omnilearn-refine.md

@@ -28,6 +28,7 @@ Both scenarios follow the same core flow: research → analyze → produce struc
 .omnilearn/<skill>/
 ├── roadmap.md
 ├── SkillPreferences.md
+├── SkillConventions.md                  ← 🔑 Setup conventions read for consistent diagnostic tasks
 ├── progress-index.md
 ├── runs/
 │   └── YYYY-MM-DD-HHMMSS-refine-<topic>/
@@ -42,12 +43,43 @@ Both scenarios follow the same core flow: research → analyze → produce struc
         └── runs/
 ```
 
+## Current Date Context
+
+```bash
+CURRENT_DATE=$(date +%Y-%m-%d)
+CURRENT_YEAR=$(date +%Y)
+```
+
+**Every subagent that does research or content generation MUST receive the current date and be told to prioritize current information over deprecated or outdated resources.**
+
+## MCP Tool Call Semantics (CRITICAL — Subagents Frequently Get This Wrong)
+
+All subagents that use MCP tools MUST follow these exact calling conventions:
+
+### `context7_resolve-library-id` + `context7_query_docs`
+1. **ALWAYS call `context7_resolve-library-id` FIRST** with the library name to get the correct library ID.
+2. Use the returned library ID (format: `/org/package`) as the `libraryId` parameter in `context7_query_docs`.
+3. Do NOT guess or hardcode library IDs.
+4. Max 3 calls per question.
+
+### `google_search` / `websearch_web_search_exa`
+- Use specific, well-formed queries — not keywords, but what you want to find.
+- Good: `"best practices for error handling in Rust 2024"`
+- Bad: `"Rust error handling"`
+- Pass the `query` parameter as a string, not an object.
+
+### General Rules
+- Match tool call parameter names EXACTLY as specified in the tool definitions.
+- Do NOT wrap string parameters in objects.
+- Do NOT nest tool calls unless required by the API.
+- If a tool call fails, check parameter names and types before retrying.
+
 ## MANDATORY TOOLS
 
 | Tool | When | Why |
 |------|------|-----|
 | `google_search` / `websearch_web_search_exa` | Research | Topic research, answer questions, validate changes |
-| `context7_query_docs` | Tech topics | Official documentation for precise answers |
+| `context7_resolve-library-id` + `context7_query_docs` | Tech topics | Official documentation for precise answers — MUST call resolve first |
 | `task(category="unspecified-high", background)` | Heavy work | Research, analysis, topic updates |
 | `read`, `write`, `edit`, `bash`, `grep`, `glob` | All phases | File operations |
 

package/commands/omnilearn-roadmap-edit.md

@@ -12,6 +12,37 @@ description: Edit an existing learning roadmap with full research-backed revisio
 /omnilearn-roadmap-edit Python I want to add web scraping and automation
 ```
 
+## Current Date Context
+
+```bash
+CURRENT_DATE=$(date +%Y-%m-%d)
+CURRENT_YEAR=$(date +%Y)
+```
+
+**All subagents that do research or content generation MUST receive the current date and prioritize current information over deprecated or outdated resources.**
+
+## MCP Tool Call Semantics (CRITICAL — Subagents Frequently Get This Wrong)
+
+All subagents that use MCP tools MUST follow these exact calling conventions:
+
+### `context7_resolve-library-id` + `context7_query_docs`
+1. **ALWAYS call `context7_resolve-library-id` FIRST** with the library name to get the correct library ID.
+2. Use the returned library ID (format: `/org/package`) as the `libraryId` parameter in `context7_query_docs`.
+3. Do NOT guess or hardcode library IDs.
+4. Max 3 calls per question.
+
+### `google_search` / `websearch_web_search_exa`
+- Use specific, well-formed queries — not keywords, but describe the ideal page.
+- Good: `"current best practices for REST API design 2025"`
+- Bad: `"REST API"`
+- Pass `query` as a plain string, not wrapped in an object.
+
+### General Rules
+- Match tool call parameter names EXACTLY as defined in the tool schema.
+- Do NOT wrap string parameters in extra objects or arrays.
+- Do NOT nest tool calls unless the API explicitly requires it.
+- If a tool call fails, verify parameter names match before retrying.
+
 ## Core Behavior
 
 This command performs the same rigorous research-backed process as `/omnilearn-roadmap`, but operates on an EXISTING roadmap. It:
@@ -32,6 +63,7 @@ This command performs the same rigorous research-backed process as `/omnilearn-r
 ├── UserPreferences.md
 └── <skill-name>/
     ├── SkillPreferences.md
+    ├── SkillConventions.md                  ← Preserved during edits
     ├── roadmap.md                          ← Will be updated
     ├── progress-index.md                   ← Must be preserved & updated
     ├── runs/
@@ -228,6 +260,7 @@ task(category="unspecified-high", run_in_background=true, prompt="
 3. REQUIRED TOOLS: google_search, websearch_web_search_exa, context7_resolve-library-id, context7_query_docs, read, write
 
 4. MUST DO:
+   - **CURRENT DATE: {CURRENT_DATE}** — ONLY research current information. Check for deprecation warnings. Prioritize resources from the last 1-2 years. Prefer latest stable versions of any library/framework/tool.
    - Analyze the user's change request: '{user_change_request}'
    - Use google_search / websearch_web_search_exa to research:
      * The specific topics/areas the user wants to add or modify

package/commands/omnilearn-roadmap.md

@@ -19,6 +19,7 @@ description: Create a comprehensive, real-world-ready learning roadmap for any s
 ├── UserPreferences.md                           ← Global user preferences (auto-read + updated)
 └── <skill-name>/                                 ← Normalized skill folder (e.g., "Rust", "Machine-Learning")
     ├── SkillPreferences.md                       ← Per-skill learning preferences
+    ├── SkillConventions.md                       ← 🔑 Setup conventions: package manager, project structure, deps, testing (auto-learned)
     ├── roadmap.md                                ← Master roadmap (the main deliverable)
     ├── progress-index.md                         ← Progress log — index of all runs
     ├── runs/                                     ← Research & execution logs
@@ -88,6 +89,37 @@ echo "Using learning directory: $LEARNING_DIR"
 
 All subsequent paths in this command use `$OMNILEARN_DIR` as the base.
 
+## Current Date Context (CRITICAL — Must Pass to ALL Subagents)
+
+```bash
+CURRENT_DATE=$(date +%Y-%m-%d)
+CURRENT_YEAR=$(date +%Y)
+```
+
+**Every subagent that does research or content generation MUST receive the current date and be told to prioritize current information.**
+
+## MCP Tool Call Semantics (CRITICAL — Subagents Frequently Get This Wrong)
+
+All subagents that use MCP tools MUST follow these exact calling conventions:
+
+### `context7_resolve-library-id` + `context7_query_docs`
+1. **ALWAYS call `context7_resolve-library-id` FIRST** with the library name to get the correct library ID.
+2. Use the returned library ID (format: `/org/package`) as the `libraryId` parameter in `context7_query_docs`.
+3. Do NOT guess or hardcode library IDs.
+4. Max 3 calls per question.
+
+### `google_search` / `websearch_web_search_exa`
+- Use specific, well-formed queries — not keywords, but describe the ideal page.
+- Good: `"blog post comparing Rust error handling patterns 2024"`
+- Bad: `"Rust error handling"`
+- Pass `query` as a plain string, not wrapped in an object.
+
+### General Rules
+- Match tool call parameter names EXACTLY as defined in the tool schema.
+- Do NOT wrap string parameters in extra objects or arrays.
+- Do NOT nest tool calls unless the API explicitly requires it.
+- If a tool call fails, verify parameter names match before retrying.
+
 ## MANDATORY TOOLS
 
 | Tool | When | Why |
@@ -196,6 +228,7 @@ task(category="unspecified-high", run_in_background=true, prompt="
 3. REQUIRED TOOLS: google_search, websearch_web_search_exa, context7_resolve-library-id, context7_query_docs, read, write
 
 4. MUST DO:
+   - **CURRENT DATE: {CURRENT_DATE}** — ONLY search for and reference current information. Prioritize resources from the last 1-2 years. Check for deprecation warnings on any library/framework/tool. Prefer latest stable versions. If something has been superseded (e.g. Create React App → Vite, npm → pnpm/uv), flag it.
 
    FIRST — Consider the user's level and existing knowledge:
    - User's stated experience with {skill}: {user level}
@@ -255,6 +288,7 @@ task(category="unspecified-high", run_in_background=true, prompt="
 3. REQUIRED TOOLS: google_search, websearch_web_search_exa, read, write
 
 4. MUST DO:
+   - **CURRENT DATE: {CURRENT_DATE}** — ONLY search for and reference current information. Prioritize resources from the last 1-2 years. Check for deprecation warnings on any library/framework/tool. Prefer latest stable versions. If something has been superseded (e.g. Create React App → Vite, npm → pnpm/uv), flag it.
 
    FIRST — Consider the user's level:
    - User's experience with {skill}: {user level}
@@ -336,11 +370,12 @@ task(category="unspecified-high", run_in_background=false, timeout=300000, promp
 3. REQUIRED TOOLS: google_search, websearch_web_search_exa, context7_query_docs, read, write, grep
 
 4. MUST DO:
+   - **CURRENT DATE: {CURRENT_DATE}** — ONLY reference current information. If you do additional online research, prioritize resources from the last 1-2 years. Check for deprecation warnings. Prefer latest stable versions of any library/framework/tool mentioned.
    - Read BOTH research files:
      * Content research: {RUNS_DIR}/research-content.md
      * Learning path research: {RUNS_DIR}/research-learning.md
    - Read cross-skill context: existing skills the user has
-   - If there are gaps or contradictions, do additional research online
+   - If there are gaps or contradictions, do additional research online (using current date context)
    
    CRITICAL — Personalization rules:
    - **Do NOT include topics the user already knows.** If the user is experienced in Python and learning FastAPI, skip Python fundamentals entirely.

package/commands/omnilearn-start.md

@@ -30,7 +30,8 @@ The goal metric: **Can the user apply what they learned to real-world problems t
 .omnilearn/<skill>/
 ├── roadmap.md
 ├── SkillPreferences.md
-├── progress-index.md                       ← Overview index: links to topic-progress.md per topic
+├── SkillConventions.md                  ← 🔑 Setup conventions: package manager, project structure, deps, testing preferences (auto-learned)
+├── progress-index.md                   ← Overview index: links to topic-progress.md per topic
 ├── runs/                                   ← Skill-level run logs (action logs only, NOT progress)
 │   └── YYYY-MM-DD-HHMMSS-learning-<topic>/
 │       ├── agent-log.md                    ← What happened this run: decisions, actions taken
@@ -57,6 +58,37 @@ The goal metric: **Can the user apply what they learned to real-world problems t
                 └── ...
 ```
 
+## Current Date Context (CRITICAL — Must Pass to ALL Subagents)
+
+```bash
+CURRENT_DATE=$(date +%Y-%m-%d)
+CURRENT_YEAR=$(date +%Y)
+```
+
+**Every subagent that does research or content generation MUST receive the current date and be told to prioritize current information over deprecated or outdated resources.**
+
+## MCP Tool Call Semantics (CRITICAL — Subagents Frequently Get This Wrong)
+
+All subagents that use MCP tools MUST follow these exact calling conventions:
+
+### `context7_resolve-library-id` + `context7_query_docs`
+1. **ALWAYS call `context7_resolve-library-id` FIRST** with the library name to get the correct library ID.
+2. Use the returned library ID (format: `/org/package`) as the `libraryId` parameter in `context7_query_docs`.
+3. Do NOT guess or hardcode library IDs.
+4. Max 3 calls per question.
+
+### `google_search` / `websearch_web_search_exa`
+- Use specific, well-formed queries — not keywords, but describe the ideal page.
+- Good: `"current state of Rust async patterns 2025"`
+- Bad: `"Rust async"`
+- Pass `query` as a plain string, not wrapped in an object.
+
+### General Rules
+- Match tool call parameter names EXACTLY as defined in the tool schema.
+- Do NOT wrap string parameters in extra objects or arrays.
+- Do NOT nest tool calls unless the API explicitly requires it.
+- If a tool call fails, verify parameter names match before retrying.
+
 ## MANDATORY TOOLS
 
 | Tool | When | Why |
@@ -111,11 +143,12 @@ if [ ! -d "$SKILL_DIR" ] || [ ! -f "$ROADMAP" ]; then
 fi
 ```
 
-### 0.3 Read Progress, Preferences & Cross-Skill Inventory
+### 0.3 Read Progress, Preferences, Conventions & Cross-Skill Inventory
 
 Read these files to understand the current state:
 - `UserPreferences.md` (if exists) — learning style, experience level, goals
 - `SkillPreferences.md` — per-skill state, preferences
+- **`SkillConventions.md`** (if exists) — 🔑 **Setup conventions: package manager, project structure, deps, testing preferences. Read this BEFORE generating any scaffold or assignment — it ensures consistency.**
 - `progress-index.md` — overview of topic statuses
 
 **For any topics marked 🔵 In Progress**, read their `topic-progress.md` files to get the detailed state:
@@ -181,6 +214,7 @@ task(category="deep", run_in_background=false, timeout=600000, prompt="
 3. REQUIRED TOOLS: google_search, websearch_web_search_exa, context7_resolve-library-id, context7_query_docs, read, write, bash, grep
 
 4. MUST DO — Step-by-Step:
+   - **CURRENT DATE: {CURRENT_DATE}** — ONLY research current information. Prioritize resources from the last 1-2 years. Check for deprecation warnings. Prefer latest stable versions of any library/framework/tool. If older resources reference superseded tools (e.g. CRA → Vite, pip → uv), flag and use the current standard.
 
    STEP 1 — Research:
    - Read the main roadmap to understand how this topic fits: {ROADMAP}
@@ -293,6 +327,7 @@ task(category="deep", run_in_background=false, timeout=600000, prompt="
    - Topics directory: {TOPICS_DIR}/{topic}/
    - User experience level: {from preferences}
    - User learning style: {from preferences}
+   - Skill conventions (package manager, project structure, deps, testing setup): {from SkillConventions.md — read before generating scaffold}
 ")
 ```
 
@@ -369,6 +404,7 @@ task(category="unspecified-high", run_in_background=false, prompt="
 3. REQUIRED TOOLS: google_search, websearch_web_search_exa, context7_query_docs, read, write
 
 4. MUST DO:
+   - **CURRENT DATE: {CURRENT_DATE}** — ONLY research current information. Check for deprecation warnings. If a library/framework has a newer standard (e.g. CRA → Vite, pip → uv), reference the current approach.
    - Read the current assignment: {ASSIGNMENT_DIR}/question.md (to understand context without giving away the solution)
    - Research the concept online to find the MOST COMMON confusion points
    - Design a micro-exercise that:
@@ -441,6 +477,7 @@ task(category="writing", run_in_background=false, prompt="
 3. REQUIRED TOOLS: read, write
 
 4. MUST DO:
+   - **CURRENT DATE: {CURRENT_DATE}** — Use current, real-world examples. Avoid outdated patterns or deprecated APIs.
    - Read the current assignment to understand what's been covered
    - Create 2-3 smaller exercises that target the specific area the user is struggling with
    - Each should be solvable in 5-15 minutes
@@ -452,6 +489,8 @@ task(category="writing", run_in_background=false, prompt="
 ")
 ```
 
+```
+
 ### 3.2 Track All Interactions
 
 For each user interaction during the session:
@@ -499,6 +538,7 @@ task(category="unspecified-high", run_in_background=false, timeout=300000, promp
 3. REQUIRED TOOLS: google_search, websearch_web_search_exa, context7_query_docs, read, write
 
 4. MUST DO:
+   - **CURRENT DATE: {CURRENT_DATE}** — ONLY use current, real-world scenarios. Avoid outdated APIs, deprecated libraries, or superseded best practices. Research what's current in the industry for this topic.
    - Read the topic-roadmap.md: {TOPICS_DIR}/{topic}/topic-roadmap.md
    - Read the previous assignment(s) to ensure progression: {ASSIGNMENT_DIR}
    - Read the user's learning preferences and history from {SKILL_PREFS}
@@ -547,6 +587,7 @@ task(category="unspecified-high", run_in_background=false, timeout=300000, promp
    - Difficulty: {basic/intermediate/real-world}
    - Previous assignment concepts: {summary}
    - User performance on previous assignment: {observations}
+   - Skill conventions (package manager, project structure, deps, testing setup): {from SkillConventions.md — MUST follow these when generating scaffold}
 ")
 ```
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnilearn-workflow",
-  "version": "1.0.0",
+  "version": "1.0.2",
   "description": "OmniLearn — AI-powered adaptive learning workflow for OpenCode. Multi-agent orchestration that creates personalized roadmaps, hands-on assignments, and tracks progress across any skill.",
   "keywords": [
     "opencode",