compound-learnings
Compound Learnings
You are capturing knowledge from the work just completed so that future sessions in this project are smarter. This is the compound interest of engineering — each session makes the next one better.
When to Activate
- Invoked by the
shipcommand after PR creation and Linear update - After any significant work session, even if not formally shipping
- NOT after trivial changes (typos, version bumps, single-line fixes)
Preconditions
Before compounding, validate inputs exist:
- Diff exists: Detect the base branch first:
base_branch=$(git symbolic-ref refs/remotes/origin/HEAD 2>/dev/null | sed 's|refs/remotes/origin/||' || echo main), then rungit log "$base_branch"..HEAD --onelinevia Bash (this is a git command, not file content). If the output is empty, skip compounding with: "No commits on branch. Nothing to compound." - CLAUDE.md exists: Use the Read tool to read the project root CLAUDE.md. If missing, ask the developer via AskUserQuestion: "No CLAUDE.md found. Create one with
/workflows:setup-claude-md, or skip compounding?"
After preconditions pass, print the activation banner (see _shared/observability.md):
More from brite-nites/britenites-claude-plugins
verification-before-completion
Ensures tasks are genuinely resolved before marking them done. Activates at task checkpoints during plan execution — validates that fixes actually work, tests genuinely pass, and acceptance criteria are met. Prevents premature completion declarations.
16writing-plans
Breaks work into bite-sized tasks before coding. Activates when a multi-step task needs planning — creates tasks small enough for a junior developer to follow (2-5 minutes each), with exact file paths, complete implementation details, and verification steps. References Linear issue context and project-specific test commands from CLAUDE.md.
14systematic-debugging
Four-phase root cause analysis for bug investigation. Activates when debugging unexpected behavior, failing tests, or production issues — follows reproduce, isolate, analyze, fix with defense-in-depth. Uses condition-based waiting instead of arbitrary delays. Available anytime, not tied to the inner loop sequence.
14refine-plan
Refines a v1 project plan into agent-ready tasks with clear context, implementation steps, and validation criteria. Use after /plan-project has produced a v1 plan.
13setup-claude-md
Generates a best-practices CLAUDE.md file for the project. Analyzes the codebase and applies Claude Code best practices for optimal agent performance. Use at project setup or after /create-issues.
13executing-plans
Executes a structured plan using subagent-per-task with TDD enforcement. Activates when given an approved plan to implement — launches fresh subagents for each task, enforces red-green-refactor, runs two-stage review per task, and checkpoints between tasks. Parallelizes independent tasks.
13