boundary-guardian — Claude Code subagent: setup and details
boundary-guardian monitors reasoning boundaries and detects semantic drift to stop AI hallucinations before invalid statements occur.
no license ★ 1,357 updated 2026-03-01
.claude/agents/boundary-guardian.mdWhat boundary-guardian does
The Sentinel subagent assigns reasoning steps into four color-coded categories based on a tension metric (ΔS): safe, transitional, risk, and danger. Crossing into the danger zone prompts an immediate halt to reasoning and triggers a BBCR sequence to retreat to verified concepts.
It provides built-in commands prefixed with an asterisk, such as *detect-boundary, *assess-risk, and *show-heatmap. Because its command execution resolves against markdown scripts inside .claude/commands/boundary/ and .claude/commands/wfgy/, the agent cannot run those specific procedures if those directories are missing.
What it helps with
- detect when reasoning drifts past verifiable knowledge boundaries
- stop runaway hallucinations using an automated BBCR fallback sequence
- evaluate conversation risk levels across categorized metric zones
- build bridging concepts to explore uncertain topics without guessing
Example prompts for the agent
- Run *detect-boundary to inspect current knowledge limits
- Check the risk score of this explanation with *assess-risk
- Build a safe conceptual path toward the target topic using *find-bridge
How to install boundary-guardian
Manually: put the files in place
.claude/agents/boundary-guardian.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/boundary-guardian.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/boundary-guardian.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 boundary-guardian
How to install boundary-guardian in Claude Code?
Add the agent definition file to your workspace at .claude/agents/boundary-guardian.md.
What does boundary-guardian do?
It tracks semantic tension thresholds to identify when model reasoning approaches unreliable territory and enforces corrective recovery steps.
What dependencies does this subagent require?
It expects related command files to exist in .claude/commands/boundary/ and .claude/commands/wfgy/ when executing specific operations.