# Compound

> Capture one compounding repo-technical learning while a solved problem is still fresh. Use when: after a bug fix, diagnosis, delivery, review, or incident reveals a reusable pattern worth adding to `docs/solutions/`. Trigger: /compound, /capture-learning, /learning.

- Skill: `phrazzld/compound` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add phrazzld/compound`
- Raw SKILL.md: https://api.skillmd.com/api/skills/phrazzld/compound/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: phrazzld (https://skillmd.com/u/phrazzld)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/phrazzld/compound

---


# /compound

Capture the learning, not the whole story.

## Contract

- One solved problem -> one learning.
- Corpus: `docs/solutions/<category>/<slug>.md`; schema:
  `harnesses/shared/references/learnings.md`.
- Overlap first:
  `rg -n --glob '*.md' '^(title|tags|applies_when):|<module>|<failure-mode>' docs/solutions`.
- If an existing learning covers it, update that file only when the fresh
  evidence contradicts or sharpens it. Otherwise do not add a near-duplicate.
- If new, use grep-ready frontmatter: `title`, `tags`, `module`,
  `problem_type`, `applies_when`, `severity`, `date`.
- Cite checkable evidence: `repo@SHA`, PR, command, route, and `file:line`
  where possible. Session notes may provide context; they are not the anchor.
- Pattern docs are promoted from multiple learnings. Never author a broad
  doctrine page cold from one bug.

## Write Discipline

Present evidence wins at read. Coherence-neighborhood at write: refresh only
nearby learnings likely to conflict. On contradiction, scope the refresh to
the smallest corpus slice that can be wrong.

## Output

Report the path written or updated, the overlap query used, the evidence
anchor, and 3-6 retrieval terms future agents should grep.