semantic-architect — Claude Code subagent: setup and details
semantic-architect is a Claude Code subagent that organizes dialogue context into tree structures and preserves persistent memory across sessions.
no license ★ 1,357 updated 2026-03-01
.claude/agents/semantic-architect.mdWhat semantic-architect does
Operating under the persona Atlas, semantic-architect structures conversational takeaways into connected hierarchical branches inside the WFGY setup. On launch, it restores context from the local .wfgy/trees/ folder and triggers initialization routines. Specific operations run through asterisk-prefixed directives such as *init-tree, *record-node, or *compress-tree, invoking sub-commands stored across .claude/commands/semantic/ and .claude/commands/wfgy/.
Built-in parameters limit the environment to 10 trees, capping node count at 10000 items and text length at 5000 characters per unit. The configuration triggers automatic tree compression once reaching 1000 nodes, while creating rollback checkpoints every 50 additions. Data dumps can be saved as plain text, JSON, or Markdown files.
The subagent explicitly halts for manual user confirmation before carrying out major structural edits. It depends on auxiliary command files in project directories to handle tasks like node merging and pruning.
What it helps with
- retain context and research insights across terminal sessions
- organize conversational outcomes into structured semantic trees
- export persistent memory graphs to Markdown or JSON
- prune low-value notes and merge duplicate concept nodes
Example prompts for the agent
- Display the active knowledge structure via
*view-tree - Save our current decision into a new semantic node
- Export the active reasoning tree into a markdown file
How to install semantic-architect
Manually: put the files in place
.claude/agents/semantic-architect.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/semantic-architect.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/semantic-architect.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about semantic-architect
How to install semantic-architect in Claude Code?
Place the agent definition file at .claude/agents/semantic-architect.md.
What does semantic-architect do?
It acts as a persistent memory architect, storing dialogue insights into structured, exportable node graphs.
What are the core capacity limits in this agent configuration?
It caps capacity at 10 separate trees, 10000 nodes per structure, and 5000 characters per individual node.