do:compound

/compound — Context-Lean Edition

Capture problem solutions while context is fresh. Uses parallel subagents with disk persistence.

Usage

/do:compound                    # Document the most recent fix
/do:compound [brief context]    # Provide additional context hint

Preconditions

• Problem has been solved or finding has been validated (not in-progress speculation)
• Solution has been verified working OR analysis has been confirmed against evidence
• Non-trivial insight (not simple typo or routine task)

Mode Detection

Assess whether this is a code solution or analytical/strategic finding:

Signal	Code Solution	Analytical/Strategic
Artifacts	Code changes, config fixes, test results	Documents, models, research findings, decisions
Verification	Tests pass, bug no longer reproduces	Evidence checked against primary sources, logic validated
Value	"Next time this breaks, do X"	"Next time this question comes up, here's what we found"

If analytical/strategic: Use the modified agent prompts and doc template described below. The same Phase 1->2->3 structure applies, but agents focus on different things.

Execution: Two-Phase Orchestration

Phase 1: Parallel Research (Disk-Persisted)

Derive a short topic stem from the problem or finding being compounded (e.g., redis-cache-invalidation or api-versioning-strategy). Use lowercase, hyphens, 3-6 words max.

Resolve main root for .workflows/ paths: Run git worktree list --porcelain | head -1 | sed 's/^worktree //' to get the main repo root. Set WORKFLOWS_ROOT=<main-root>/.workflows. All .workflows/ paths in this skill use $WORKFLOWS_ROOT, NOT relative .workflows/. This ensures artifacts survive worktree lifecycle transitions and are shared across sessions.

mkdir -p $WORKFLOWS_ROOT/compound-research/<topic-stem>/agents

Launch 5 subagents in parallel with run_in_background: true:

1. Context Analyzer

Task general-purpose (run_in_background: true): "
You are a context analyzer specializing in problem classification and root cause identification. Analyze the conversation history. Identify problem type, component, symptoms.
Validate against schema categories: build-errors, test-failures, runtime-errors,
performance-issues, database-issues, security-issues, ui-bugs, integration-issues, logic-errors.

=== OUTPUT INSTRUCTIONS (MANDATORY) ===
Write YAML frontmatter skeleton to: $WORKFLOWS_ROOT/compound-research/<topic-stem>/agents/context.md
Return ONLY a 2-3 sentence summary.
"

2. Solution Extractor

Task general-purpose (run_in_background: true): "
You are a solution extractor specializing in distilling actionable knowledge from problem-solving sessions. Extract the key solution, evidence chain, alternatives considered and why they were rejected, and the decision rationale.

=== OUTPUT INSTRUCTIONS (MANDATORY) ===
Write findings to: $WORKFLOWS_ROOT/compound-research/<topic-stem>/agents/solution.md
Return ONLY a 2-3 sentence summary.
"

3. Related Docs Finder

Task general-purpose (run_in_background: true): "
You are a documentation researcher specializing in connecting new findings to existing institutional knowledge. Find related brainstorms, plans, solutions, and prior analyses. Map how this finding connects to existing knowledge.

=== OUTPUT INSTRUCTIONS (MANDATORY) ===
Write findings to: $WORKFLOWS_ROOT/compound-research/<topic-stem>/agents/related-docs.md
Return ONLY a 2-3 sentence summary.
"

4. Prevention Strategist

Task general-purpose (run_in_background: true): "
You are a prevention strategist specializing in root cause analysis and recurrence prevention. Identify: How could this have been caught earlier? What systemic changes would prevent recurrence? What monitoring or tests should be added?

=== OUTPUT INSTRUCTIONS (MANDATORY) ===
Write findings to: $WORKFLOWS_ROOT/compound-research/<topic-stem>/agents/prevention.md
Return ONLY a 2-3 sentence summary.
"

5. Category Classifier

Task general-purpose (run_in_background: true): "
You are a knowledge classifier specializing in taxonomizing technical findings for future retrieval. Classify the problem type and determine the best category directory for the solution document.

=== OUTPUT INSTRUCTIONS (MANDATORY) ===
Write findings to: $WORKFLOWS_ROOT/compound-research/<topic-stem>/agents/category.md
Return ONLY a 2-3 sentence summary.
"

Monitor completion via file existence. DO NOT call TaskOutput.

Analytical/Strategic Mode — Agent Adaptations

When in analytical/strategic mode, the same 5 agents run but with adapted focus:

1. Context Analyzer — Identify the question/problem type, domain area, and what triggered the investigation. Categories include: competitive-intelligence, cost-modeling, organizational-dynamics, technical-evaluation, strategic-positioning, data-analysis, stakeholder-management.
2. Solution Extractor — Extract the key finding, the evidence chain that supports it, alternatives that were considered and why they were rejected, and the decision rationale.
3. Related Docs Finder — Find related brainstorms, plans, meeting transcripts, and prior analyses. Map how this finding connects to existing knowledge.
4. Prevention Strategist -> Reuse Strategist — Instead of "how to prevent recurrence," capture: When would this analysis be relevant again? What triggers should prompt re-reading this? What assumptions could invalidate the finding?
5. Category Classifier — Same role, broader categories (see list in #1 above).

Phase 2: Assembly & Write

After all Phase 1 agents complete:

1. Read all files from $WORKFLOWS_ROOT/compound-research/<topic-stem>/agents/
2. Assemble complete markdown file from the pieces
3. Validate YAML frontmatter against schema. Include traceability fields:

   origin_plan: docs/plans/YYYY-MM-DD-<name>-plan.md      # if traceable
   origin_brainstorm: docs/brainstorms/YYYY-MM-DD-<name>-brainstorm.md  # if traceable
   

   Trace the origin chain: check recent git log, beads issues, or conversation context for the plan and brainstorm that led to this solution. If not traceable, omit the fields.
4. Create directory if needed: mkdir -p docs/solutions/[category]/
5. Write the SINGLE final file: docs/solutions/[category]/[filename].md

Phase 3: Optional Enhancement

Based on problem type, optionally run specialized review agents:

Task performance-oracle (run_in_background: true): "You are a performance specialist. Review the solution for performance implications, scalability concerns, and optimization opportunities.
Read the solution doc at: docs/solutions/[category]/[filename].md
=== OUTPUT INSTRUCTIONS (MANDATORY) ===
Write findings to: $WORKFLOWS_ROOT/compound-research/<topic-stem>/agents/perf-review.md
Return ONLY a 2-3 sentence summary."

Task security-sentinel (run_in_background: true): "You are a security reviewer. Review the solution for security implications, vulnerability exposure, and hardening opportunities.
Read the solution doc at: docs/solutions/[category]/[filename].md
=== OUTPUT INSTRUCTIONS (MANDATORY) ===
Write findings to: $WORKFLOWS_ROOT/compound-research/<topic-stem>/agents/security-review.md
Return ONLY a 2-3 sentence summary."

Task data-integrity-guardian (run_in_background: true): "You are a data integrity specialist. Review the solution for data consistency, migration safety, and schema implications.
Read the solution doc at: docs/solutions/[category]/[filename].md
=== OUTPUT INSTRUCTIONS (MANDATORY) ===
Write findings to: $WORKFLOWS_ROOT/compound-research/<topic-stem>/agents/data-review.md
Return ONLY a 2-3 sentence summary."

Run only the agents relevant to the problem type. Read output files and incorporate significant findings into the solution doc.

Retain Research

Do NOT delete research outputs. The research directory at $WORKFLOWS_ROOT/compound-research/<topic-stem>/ is retained for traceability and learning.

What It Captures

Code Solutions

• Problem symptom (exact error messages)
• Investigation steps tried
• Root cause analysis
• Working solution with code examples
• Prevention strategies
• Cross-references to related docs

Analytical/Strategic Findings

• Question or problem that triggered the investigation
• Evidence chain (data, transcripts, documents, external sources)
• Key finding and why it matters
• Decision rationale — what was chosen and what was rejected
• Assumptions that could invalidate the finding
• Reuse triggers — when to re-read this (e.g., "next time pricing comes up", "before any vendor evaluation meeting")
• Stakeholder implications — who is affected and how
• Cross-references to brainstorms, plans, meeting transcripts

Output

Single file: docs/solutions/[category]/[filename].md

The Compounding Philosophy

Each documented solution compounds knowledge:

1. First occurrence: Research (30 min)
2. Document: 5 min
3. Next occurrence: Quick lookup (2 min)

Each unit of engineering work should make subsequent units easier — not harder.