compound-engineering

Targeted gap analysis to push a project forward — find what matters most and do it next

Check or stop the post-work chain (work → review → discover)

Implement discovered patterns into the local project plugin and update subsystem knowledge

Craft a comprehensive prompt for a frontier reasoning model — clarify intent, gather context, assemble a non-prescriptive prompt

Assemble a self-contained revision prompt from a writing contract, numbered draft, and annotations — for single-shot revision in a powerful external model

Turn an approved direction into harness-ready Linear issues — self-bootstrapping bridge from investigation/reasoning to autonomous execution

Analyze work session to discover reusable patterns and specify appropriate components for the local project plugin

Systematically explore and document a subsystem for the compound engineering knowledge base

Detect and clean up accumulated knowledge debt in subsystem specs, discovery artifacts, and harness state

Orchestrate autonomous engineering — delegates each issue to a sub-agent with fresh context for plan→work→review→commit

Interactive investigation of architecture and subsystems — visual mental models, deep dives, gap analysis, and proposal evaluation

Get context on a Linear project and issue, review previous work, and understand codebase state

Initialize orchestrator mode — explore codebase, build knowledge base, activate interactive orchestration

Produce a dependency-aware change plan by consulting subsystem knowledge before reading source code

Assemble codebase context for frontier reasoning models and transform their responses into actionable work

Verify work against subsystem contracts, invariants, and tests

Final wrap-up — verify subsystem spec, set Linear issue to Done with implementation comment, commit task-specific files

Decompose a product vision into a Linear hierarchy (initiative, project, milestones, issues) grounded in subsystem knowledge

Execute planned work while leveraging subsystem knowledge and applying the improvement mindset

This skill should be used when the user invokes "/consolidate", asks to "process discoveries", "implement compound improvements", "review pending patterns", or wants to turn discovered patterns into actual project plugin components and subsystem knowledge updates. Provides methodology for reviewing, approving, and implementing discovery artifacts.

This skill should be used when the user invokes "/discover", asks to "find patterns", "extract workflows", "what could be a skill", "what could be a command", "modularize this", "discover repeatable procedures", or wants to identify engineering patterns that could become new plugin components. Provides methodology for subsystem-aware pattern extraction and component type selection.

This skill should be used when the user asks to "assess progress", "find gaps", "what should I work on next", "push the project forward", "gap analysis", "unblock the project", or when deciding what work matters most toward a goal. Works for any domain — engineering, product, operations, content, strategy. Provides the thinking methodology for pragmatic gap identification and prioritization.

This skill should be used when running the autonomous engineering harness (/harness start), when the user asks about "harness gate rules", "autonomous work protocol", "harness failure handling", "delegation protocol", "orchestrator model", or when Claude needs guidance on autonomous decision-making during orchestrated plan-work-review-commit cycles.

This skill should be used when starting any significant engineering task, when the user asks to "work with improvement mindset", "apply compound thinking", "set up improvement cycle", or when Claude should proactively apply the improvement lens during /work execution. Provides the active thinking framework to use DURING work, not just after.

This skill should be used when the user asks about "knowledge debt", "spec staleness", "cleanup subsystem specs", "stale discoveries", or when Claude notices during work that subsystem specs seem outdated, hot-patch comments are accumulating, or friction logs are large. Provides awareness of knowledge system decay patterns.

This skill should be used after /orchestrate init activates orchestrator mode, when the user asks to "orchestrate work", "manage this project", "delegate tasks", "what should we work on", "assess the project", or when Claude is operating as an interactive orchestrator managing sub-agents and accumulating project knowledge. Also applies when the user mentions "multi-agent delegation", "sub-agent prompting", "knowledge distillation", or "ORCHESTRATOR.md".

This skill should be used when the user wants to "create a reasoning prompt", "craft a prompt for o3/R1/Gemini", "prepare a deep analysis prompt", "generate a thinking prompt", "write a prompt for a reasoning model", or needs help formulating comprehensive prompts for frontier reasoning models that give the model space to discover insights rather than confirming prescribed conclusions.

This skill should be used when the user wants to "reason about architecture", "get strategic analysis", "use a reasoning model", "prepare context for deep think", "transform reasoning output", or when Claude needs to assemble codebase context for frontier reasoning models and transform their responses into actionable work items.

This skill should be used when the user asks to "reflect on this session", "capture learnings", "what did we learn", "debrief this work", or wants to create a structured learning artifact from recent engineering work. Provides methodology for subsystem-aware reflection and structured output.

This skill should be used when the user wants to "plan a migration", "build feature X", "break down a project", "create a roadmap", "plan this initiative", "what would it take to build X", "decompose this into issues", or when Claude needs to decompose a product vision into a Linear hierarchy (initiative, project, milestones, issues) grounded in subsystem knowledge.