Stats
Actions
Tags
From recursio
Wave execution rules — agent constraints, wave barriers, review-fix evaluation, language awareness.
How this skill is triggered — by the user, by Claude, or both
Slash command
/recursio:recursive-executionThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
These apply to ALL agents in the recursive tree:
These apply to ALL agents in the recursive tree:
isolation: "worktree". Worktree agents work in a copy — files get discarded on cleanup.run_in_background: true. Background dispatch triggers worktree isolation. Use foreground Agent calls — send parallel agents in a single message for concurrency.recursive-implementer scoped to affected files with findings in the prompt.Waves execute sequentially. Within each wave, agents run in parallel.
Wave barrier: all agents must complete before running tests or starting the next wave. Send all wave agents as foreground calls in a single message when possible. Sequential dispatch is acceptable when needed.
Typical structure:
recursive-implementer)recursive-implementer agents)Adapting mid-flight: the orchestrator may add missing deps, adjust types, reorder/merge modules, fix structural issues between waves, and adjust scope boundaries.
After each wave, spawn a recursive-reviewer scoped to the wave's output:
Post-Wave 0: review contracts for richness — are interfaces complete enough for downstream modules? Fix gaps directly.
Post-implementation waves: review for bugs, structural artifacts, cross-module consistency. Fix Critical findings via implementer agents (max 2 rounds). Note Important findings.
Post-final wave: full codebase review. Fix Critical. Verify README.md and CHANGELOG.md exist — create via implementer if missing.
Report: total tests, files, max nesting depth, findings summary (found/fixed).
The wave plan's Language & Conventions section determines:
pytest, npx vitest run, cargo test)npm install, uv sync)Include Language & Conventions in every subagent prompt.
Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub jartan-llc/grimoire --plugin recursio