feat(claude-plugin): rewrite atomic-agents plugin as v2.0.0 skills-only pack - #236
Merged
Conversation
…ly pack Full rewrite of the Claude Code plugin to match 2026 plugin conventions. Subagents and commands are replaced by three skills with progressive-disclosure reference files, matching Anthropic's skill-authoring best practices. - Retire 4 subagents (architect, explorer, reviewer, schema-designer) and 3 slash commands (/atomic-create, /atomic-agent, /atomic-tool). - Introduce skills/framework (auto-triggered) with 11 reference files covering schemas, agents, tools, context providers, prompts, orchestration, memory, hooks, providers, project structure, and testing. - Introduce skills/review (auto-triggered) with a framework-specific code review checklist. - Introduce skills/new-app (slash-only via /atomic-agents:new-app) for scaffolding a fresh project. - Correct framework imports (top-level atomic_agents package), PEP 695 generic BaseTool pattern, MCPTransportType.HTTP_STREAM, ChatHistory.load instance-method semantics, TokenCountResult seven fields, run_async (not arun), and all five hook events. - Bump plugin.json and marketplace.json to 2.0.0; update marketplace description to reflect the new skills layout. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Review is a canonical subagent use case per Anthropic's docs and every one of their own review plugins (pr-review-toolkit, feature-dev, code-review). Skills-only review has structural limits: can't restrict to read-only tools, can't isolate the file-exploration load from the parent context, can't run parallel review lenses, and can't route to a different model. My v2.0.0 rewrite overcorrected by removing all subagents; the reviewer should have stayed. - Delete skills/review/, add agents/atomic-reviewer.md with read-only tools (Glob, Grep, LS, Read, NotebookRead, TodoWrite) matching Anthropic's canonical `code-reviewer` example. - Rewritten checklist for the 2026 API: top-level imports, PEP 695 BaseTool generics, MCPTransportType.HTTP_STREAM, ChatHistory.load instance-method semantics, all five hook events, run_async not arun. - framework SKILL.md now instructs Claude to delegate to the atomic-reviewer subagent via Task instead of reviewing in-thread. - README, CHANGELOG, plugin.json, marketplace.json updated to describe the hybrid skills + subagent pattern (mirrors Anthropic's plugin-dev). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Exploration fits the subagent rubric for the same reasons review does:
read-heavy work where the output is a summary, not artifacts the user
wants to iterate on in the parent thread. When a codebase has 15+
agent/tool/schema files, the main thread loading all of them to build an
architecture map is the exact context-pollution failure mode subagents
exist to prevent. Mirrors Anthropic's code-explorer in feature-dev.
- agents/atomic-explorer.md with read-only tools (Glob, Grep, LS, Read,
NotebookRead, TodoWrite), sonnet model, yellow color. Targeted
framework-aware greps (BaseIOSchema, AtomicAgent[, BaseTool[,
BaseDynamicContextProvider, register_hook, fetch_mcp_tools) drive
component discovery. Output is a structured map with file:line
references and an essential-reading list, bounded to ~two screens.
- framework SKILL.md now instructs Claude to delegate exploration
("explore", "map", "understand how X works") to the explorer subagent
for non-trivial projects, and to read directly for small ones.
- README, CHANGELOG, plugin.json, marketplace.json updated to describe
the two-subagent hybrid.
Not added: architect / schema-designer / generator subagents. Those
produce artifacts the user wants in-thread to iterate on — anti-pattern
for subagent isolation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
claude-plugin/atomic-agentsto match 2026 Claude Code plugin conventions: skills-only, no subagents, nocommands/, progressive disclosure viareferences/.MCPTransportType.HTTP_STREAM,ChatHistory.loadsemantics,run_async, all five hook events, sevenTokenCountResultfields).1.0.0→2.0.0in bothclaude-plugin/atomic-agents/.claude-plugin/plugin.jsonand the root.claude-plugin/marketplace.json; updates the marketplace description to reflect the new skills layout.What changed
Retired: 4 subagents (
atomic-architect,atomic-explorer,atomic-reviewer,schema-designer) and 3 slash commands (/atomic-create,/atomic-agent,/atomic-tool).Introduced:
skills/framework/— auto-triggered.SKILL.md(orientation + routing) plus 11 reference files:schemas.md,agents.md,tools.md,context-providers.md,prompts.md,orchestration.md,memory.md,hooks.md,providers.md,project-structure.md,testing.md.skills/review/— auto-triggered. Framework-specific code-review checklist.skills/new-app/— slash-only (/atomic-agents:new-app). Opinionated scaffolder.Authored to Anthropic's 2026 skill-authoring standard: third-person descriptions, SKILL.md bodies under 500 lines, references one level deep, no
README.md/CHANGELOG.mdinside skill folders, inline pointers to workingatomic-examples/directories (hooks-example,deep-research,progressive-disclosure,orchestration-agent,mcp-agent,quickstart).See
claude-plugin/atomic-agents/CHANGELOG.mdfor the full 2.0.0 entry.Test plan
claude --plugin-dir ./claude-plugin/atomic-agents/plugin validate ./claude-plugin/atomic-agentspasses/atomic-agents:new-app <name>scaffolds a runnable projectframeworkskill auto-triggers and references load on demandreviewskill auto-triggers and produces framework-specific findings2.0.0on next refresh🤖 Generated with Claude Code