Compound

Overview

Captures problem solutions while context is fresh, creating structured documentation in docs/solutions/ with YAML frontmatter for searchability. Uses parallel subagents to gather context, extract the solution, find related docs, develop prevention strategies, and classify the category — then the orchestrator assembles and writes a single final file.

<critical_requirement> Only ONE file gets written — the final documentation.

Phase 1 subagents return TEXT DATA to the orchestrator. They must NOT write files. Only the orchestrator (Phase 2) writes the final documentation file. </critical_requirement>

When to Use

Trigger phrases: "that worked", "it's fixed", "working now", "problem solved", "tests are passing"

Use when:

• Problem is solved and solution is verified working
• Solution is non-trivial (not a simple typo or obvious error)
• Context is fresh — investigation steps are in recent conversation history

Don't use when:

• Fix is still in progress or unverified
• Problem was trivial

Quick Reference

/superpowers:compound                    # Document the most recent fix
/superpowers:compound [brief context]    # Provide additional context hint
/compound --compact          # Single-pass mode for context-constrained sessions

Output location: docs/solutions/[category]/[filename].md

Categories auto-detected from problem type: build-errors · test-failures · runtime-errors · performance-issues · database-issues · security-issues · ui-bugs · integration-issues · logic-errors

Implementation

Always run full mode by default. Use compact-safe mode only when explicitly requested.

Phase 0.5: Auto Memory Scan

Before Phase 1, read MEMORY.md from the auto memory directory. If relevant entries exist for the problem being documented, pass them as a labeled supplementary excerpt to the Context Analyzer and Solution Extractor prompts. Tag any memory-sourced content incorporated into the final doc with "(auto memory [claude])".

Phase 1: Parallel Research

Launch these five subagents IN PARALLEL — each returns text data only, no files:

Subagent	Returns
Context Analyzer	YAML frontmatter skeleton (problem type, component, symptoms)
Solution Extractor	Root cause + working solution with code examples
Related Docs Finder	Cross-references, related issues, stale doc candidates
Prevention Strategist	Prevention strategies and test cases
Category Classifier	Final docs/solutions/[category]/[filename].md path

Phase 2: Assembly & Write

WAIT for all Phase 1 subagents to complete, then:

1. Collect all text results from Phase 1 subagents
2. Assemble complete markdown file
3. Validate YAML frontmatter
4. mkdir -p docs/solutions/[category]/
5. Write the single final file: docs/solutions/[category]/[filename].md

Phase 2.5: Selective Refresh

After writing, check whether older docs should be refreshed. superpowers:compound-refresh is not a default follow-up — invoke it selectively:

Condition	Action
New fix contradicts a prior doc's recommendation	Invoke superpowers:compound-refresh [doc-name]
New fix supersedes an older solution	Invoke with narrowest useful scope
Multiple related candidates in same area	Ask user whether to run targeted refresh
No related docs found, or docs still consistent	Skip

Never invoke superpowers:compound-refresh without a scope argument.

Phase 3: Optional Enhancement

After Phase 2, optionally invoke specialized agents based on problem type:

Problem type	Agent
performance_issue	performance-oracle
security_issue	security-sentinel
database_issue	data-integrity-guardian
test_failure	cora-test-reviewer
Code-heavy	kieran-rails-reviewer + code-simplicity-reviewer

Compact-Safe Mode

When context is tight: skip Phase 1 subagents entirely. In one sequential pass, extract problem/root cause/solution from conversation history (plus any relevant MEMORY.md notes), classify the category, and write a minimal doc with YAML frontmatter, 1-2 sentence problem description, root cause, key code snippets, and one prevention tip. Skip Phase 3 reviews.

Common Mistakes

❌ Wrong	✅ Correct
Subagents write files like context-analysis.md, solution-draft.md	Subagents return text data; orchestrator writes one final file
Research and assembly run in parallel	Research completes → then assembly runs
Multiple files created during workflow	Single file: docs/solutions/[category]/[filename].md