compound

Compound

Capture problem solutions while context is fresh, creating structured documentation in knowledge-base/project/learnings/ with YAML frontmatter for searchability.

Usage

Activate this skill after solving a non-trivial problem. If context about the problem is not already in the conversation, use ask_user to request a brief description.

Phase 0: Setup

Read project conventions from GEMINI.md if it exists. Apply conventions during documentation.

Branch safety check: Run run_shell_command with git branch --show-current. If the result is main or master, abort with: "Error: compound cannot run on main/master. Checkout a feature branch first."

Phase 0.5: Session Error Inventory

Before writing any learning, enumerate ALL errors encountered in this session. Output a numbered list. This step cannot be skipped.

Include:

• Wrong file paths, directories, or branch confusion
• Failed shell commands or unexpected exit codes
• API errors or unexpected responses
• Wrong assumptions that required backtracking

If genuinely no errors occurred, output: "Session error inventory: none detected."

Execution Strategy

Analyze the session context to extract the learning. Unlike the Claude Code version which spawns 5 parallel subagents, this version runs sequentially within the conversation:

1. Context Analysis

• Extract conversation history
• Identify problem type, component, symptoms
• Determine YAML frontmatter fields: title, date, category, tags

2. Solution Extraction

• Analyze investigation steps
• Identify root cause
• Extract working solution with code examples

3. Related Documentation Search

• Search knowledge-base/project/learnings/ for related documentation using grep_search
• Identify cross-references and links
• Find related GitHub issues via run_shell_command with gh issue list

4. Prevention Strategy

• Develop prevention strategies
• Create best practices guidance
• Propose enforcement: hook (strongest), skill instruction (moderate), prose rule (weakest)

5. Write Learning File

Determine the optimal knowledge-base/project/learnings/ category subdirectory. Create the file with this structure:

---
title: "<descriptive title>"
date: YYYY-MM-DD
category: <category>
tags: [tag1, tag2]
---

# Learning: <title>

## Problem

<What went wrong or what was unclear>

## Solution

<What fixed it, with code examples>

## Key Insight

<The non-obvious takeaway>

## Session Errors

<Errors from Phase 0.5 inventory, each with a **Prevention:** line>

Use write_file to create the learning document.

6. Constitution Promotion

After writing the learning, check if the insight should be promoted to a project convention:

• If the learning addresses a recurring pattern → propose adding to project conventions
• If the learning exposes a gap in rules → propose a rule addition
• Use ask_user to confirm before modifying convention files