Skip to content

coroboros/agent-skills / skills/agent-creator

agent-creator

Verified

Create or update Claude Code subagent definitions in .claude/agents/, including prompts, tools, model inheritance and permissions. Use for agent configuration or Claude Code delegation mechanics, not routine delegation in another host.

references/orchestration-patterns.md

View on GitHub
# Subagent orchestration patterns

Choose a pattern for dependencies and available tools. Inherit the user's model selection; a model-tier table is not evidence of the best allocation. Schedule work within the active host's concurrency, nesting and permission limits.

## Sequential dependencies

When analysis determines the implementation, pass its evidence to the implementer before dependent edits. A security review can proceed after the relevant artifact exists. Do independent local work while waiting; do not fabricate a downstream input.

## Independent parallel work

Split work only when each delegate has a concrete deliverable that can proceed independently. Assign disjoint file ownership where possible. Each brief carries the accepted scope, relevant inputs, user corrections and acceptance checks. Avoid simultaneous edits to shared files.

The lead continues integration, investigation or verification while delegates run. A routing-only coordinator is appropriate only when that role was explicitly chosen, not as a universal restriction.

## Hierarchy and fallback

Use nesting only when the actual tool schema and depth permit it. Current Claude Code supports bounded nesting; other hosts may expose different semantics. Omit `tools` to inherit in a Claude definition instead of writing the undocumented `tools: all`.

Without independent agents, inspect the result against the acceptance criteria and report reduced review independence. Sequential self-criticism is not fresh-context verification.

## Evidence and convergence

Give reviewers the accepted brief, actual artifact and supporting evidence. Ask them to find a counterexample; allow zero findings. Resolve disagreements by reproduction, source contracts and impact, never by a majority vote or the author's confidence score.

After a demonstrated defect is repaired, rerun the affected checks. Preserve pre-existing user changes and distinguish unrelated baseline failures. Close all accepted outcomes against the final integrated result.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall