Course Workflow

Skill: Course Workflow Orchestrator

Description

Guides course development through a 11-phase workflow. Reads current status, suggests next steps, invokes appropriate skills, updates status, and captures learnings that improve the process over time.

Instructions

Path Resolution

Resolve file paths based on your execution context:

• Plugin context (${CLAUDE_PLUGIN_ROOT} is set): Read knowledge from ${CLAUDE_PLUGIN_DATA}/knowledge/ (user data), falling back to ${CLAUDE_PLUGIN_ROOT}/knowledge/ (seed data). Write knowledge to ${CLAUDE_PLUGIN_DATA}/knowledge/.
• Clone/symlink context (default): Read and write knowledge at knowledge/ relative to the project root.

1. List Courses and Select

First, scan courses-in-progress/ for course directories. For each, read 00-course.md to get title and status.

Present a course selection:

## Courses In Progress

| # | Course | Status | Phase |
|---|--------|--------|-------|
| 1 | Andamio for Contributors | revised | 3/11 |
| 2 | Andamio for Course Creators | slts-drafted | 1/11 |
| 3 | Andamio for API Developers | slts-drafted | 1/11 |

Which course are you working on? (Or: "new" to start a new course)

If the user says "new", guide them through creating a new course directory with 01-slts.md. The 00-course.md (complete course outline) is built up through the workflow.

2. Determine Current State

Once a course is selected, read the course's 00-course.md to get:

• Current status field (in frontmatter)
• Course metadata (title, audience, etc.)

Check which workflow artifacts exist:

• 00-course.md (complete course outline — the final output)
• 01-slts.md (working file for SLT drafting/revision)
• 02-slts-quality-review.md
• 03-lesson-type-classification.md
• 04-readiness-assessment.md
• 05-delegation-map.md
• assets/screenshots/README.md (screenshot checklist)
• assets/code-examples/README.md (code example checklist)
• lessons/ directory contents

Key insight: 00-course.md is the deliverable — it contains metadata, modules, SLTs, lesson types, and assignments. 01-slts.md is the working file where SLTs are drafted and refined before being incorporated into the course outline.

3. Map Status to Next Action

| Current Status | Phase | Next Action | Skill to Invoke | |----------------|-------|-------------|-----------------| | slts-drafted | 1→2 | Assess SLT quality, apply fixes to 01-slts.md | /assess-slts | | quality-reviewed | 2→3 | Classify lesson types | /classify-lesson-types | | types-classified | 3→4 | Assess agent readiness | /self-assess-readiness | | readiness-assessed | 4→5 | Create delegation map | (manual) | | delegation-mapped | 5→6 | Gather context assets | /gather-screenshots, /gather-code-examples | | context-gathered | 6→7 | Build lessons | (lesson skills) | | building | 7 | Continue building lessons | (lesson skills) | | lessons-complete | 7→8 | Compile modules for import | /compile | | compiled | 8→9 | Run full course retrospective | /compound --rollup | | compounded | 9→10 | Move to courses/, update COURSES.md | (manual) |

Note: Phase 2 (quality review) now includes applying fixes. No separate "revise" phase — edits happen in place, patterns compound to knowledge base.

3b. Phase 6: Gather Context (Detail)

Phase 6 prepares assets needed before building lessons. The readiness assessment identifies what's missing; the gather skills create checklists for collection.

Invoke based on lesson types:

| Lesson Type | Skill | Output | |-------------|-------|--------| | Product Demo | /gather-screenshots | assets/screenshots/README.md | | Developer Documentation | /gather-code-examples | assets/code-examples/README.md | | How To Guide | (manual) | Supplementary docs in assets/ | | Exploration | (usually none) | Framing questions in lesson inputs | | Organization Onboarding | (manual) | Org-specific context in assets/ |

Workflow:

1. Run /gather-screenshots if course has Product Demo SLTs needing context
2. Run /gather-code-examples if course has Developer Documentation SLTs needing context
3. Human captures screenshots / writes code examples following checklists
4. Files saved to assets/screenshots/ or assets/code-examples/
5. Check off items in README as completed
6. When all context gathered, update status to context-gathered

Directory structure after Phase 6:

courses-in-progress/[course-slug]/
  assets/
    screenshots/
      README.md           # Capture checklist
      1.2-wallet-connect-01.png
      ...
    code-examples/
      README.md           # Code checklist
      1.2-api-auth.ts
      ...

4. Present Status and Recommendation

Output a status card:

## Course: [Course Title]

**Current Status:** [status] (Phase [n] of 11)
**Progress:** [visual indicator, e.g., ████░░░░░░ 4/11]

### Completed Artifacts
- [x] 01-slts.md
- [x] 02-slts-quality-review.md
- [ ] 03-lesson-type-classification.md
- ...

### Recommended Next Step
[Description of what to do next]

**Action:** [Specific action - invoke skill, manual step, etc.]

5. Execute or Guide

If the next step involves a skill:

• Confirm with user: "Ready to run /[skill-name]?"
• If yes, invoke the skill
• After skill completes, prompt to update status

If the next step is manual:

• Explain what needs to be done
• Offer to help with specific sub-tasks
• Prompt to update status when complete

6. Update Status

After each phase completes, update 00-course.md:

• Change status field to new value
• Add last_updated timestamp
• Add brief note about what was done

7. Capture Learnings

After each phase transition, ask:

• "Any friction in this phase?"
• "Anything that should work differently next time?"

Append learnings to workflow-learnings.md in the project root (create if doesn't exist):

## [Date] - [Course Name] - Phase [n] → [n+1]

**What happened:** [brief summary]

**Friction:** [what was hard or unclear]

**Improvement idea:** [how to make it better]

**Heuristic:** [if a generalizable pattern emerged]

7b. Compounding Knowledge

Phase 9 (mandatory): When lessons are complete, run /compound --rollup to extract all knowledge from the course before promotion. This is a required step in the workflow.

Mid-workflow (optional): You can also run compound after specific phases to capture knowledge incrementally:

| Phase Completed | Optional Command | |-----------------|------------------| | quality-reviewed (2→3) | /compound quality-review | | types-classified (4→5) | /compound classification | | readiness-assessed (5→6) | /compound readiness |

The /compound skill extracts structured knowledge from artifacts into the knowledge/ directory, which improves future skill runs.

Relationship between files:

• workflow-learnings.md = human-readable narrative, captured conversationally
• knowledge/*.yaml = machine-readable patterns, consumed by skills

8. Show Accumulated Learnings

When starting a workflow session, check workflow-learnings.md for relevant insights:

• Learnings from the same phase
• Learnings from the same course type
• Recently added heuristics

Surface these as "tips from previous runs."

Status Field Format

The status field in 00-course.md should be in YAML frontmatter:

---
course: Andamio for Contributors
status: quality-reviewed
last_updated: 2026-02-27
---

If no frontmatter exists, add it.

Workflow Learnings File

The workflow-learnings.md file accumulates insights across all courses. Structure:

# Workflow Learnings

Insights captured during course development that improve the process over time.

## Phase-Specific Learnings

### Phase 2: Assess SLT Quality
- [learning]

### Phase 4: Classify Lesson Types
- [learning]

...

## Cross-Cutting Heuristics

Patterns that apply across phases:
- [heuristic]

## Course-Specific Notes

### andamio-for-contributors
- [note]

Guidelines

• Don't skip phases. The workflow is sequential for a reason. If artifacts are missing, flag it.
• Update status promptly. Status should reflect reality. Stale status causes confusion.
• Learnings are optional but valuable. Don't force them, but always ask.
• Surface relevant history. Previous learnings make each run smarter.
• Manual steps are real steps. Guide them as carefully as automated ones.
• One course at a time. Don't try to orchestrate multiple courses in parallel.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Andamio-Platform
• Source: Andamio-Platform/coach
• License: MIT

Install and usage instructions live in the source repository linked above.