Tiny agentic loop with Docker sandbox

v1.8.0 Breaking

This release includes 1 breaking change for platform teams planning a safe upgrade.

Published 20d AI Agents & Assistants

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agentic-ai agentic-coding ai autonomous-agent cli codegolf

+6 more

coding-agent developer-tools harness llm repl sandbox

Affected surfaces

auth breaking_upgrade

ReleasePort's take

Light signal

editorial:auto 9d

The `check` parameter in Goal tool now acts as a judge prompt instead of executing a bash command. New CLI flags `-g` and `-c` enable goal loop execution modes.

Why it matters: Affects scripts using the `check` parameter; update automation to send judge prompts, not shell commands. Adopt `-g`/`-c` for loop control in version v1.8.0.

Summary

AI summary

check parameter now functions as a judge prompt rather than a bash command.

Changes in this release

Type	Severity	Summary	CVE
Breaking	Medium	Goal tool's `check` parameter now a judge prompt, not bash command. Goal tool's `check` parameter now a judge prompt, not bash command. Source: llm_adapter@2026-05-21 Confidence: low	—
Feature	Medium	`-g` and `-c` CLI flags introduced for goal loop execution mode. `-g` and `-c` CLI flags introduced for goal loop execution mode. Source: llm_adapter@2026-05-21 Confidence: high	—
Refactor	Medium	Persona adjusted to terse, unimpressed style with one-sentence caps. Persona adjusted to terse, unimpressed style with one-sentence caps. Source: llm_adapter@2026-05-21 Confidence: low	—
Refactor	Low	Persona responses are now terse, unimpressed, and limited to one-sentence caps for non-task input. Persona responses are now terse, unimpressed, and limited to one-sentence caps for non-task input. Source: granite4.1:30b@2026-05-22-audit Confidence: low	—

Full changelog

Breaking

Goal tool: check parameter is now a judge prompt, not a bash command. A mi subagent evaluates the work (can read files, run commands) and responds ACK/NACK. For bash checks, include the command in the prompt: check="run npm test — ACK if pass, NACK otherwise".

Added

-g / -c CLI flags: Goal loop as a first-class execution mode. mi -g "goal" -c "judge criteria" iterates worker/judge subagent pairs until the judge ACKs.

Changed

Persona: terse, unimpressed, one-sentence cap for non-task input. Drier responses to philosophical and off-topic questions.

Breaking Changes

The `check` parameter in Goal tool is now a judge prompt (evaluated by a subagent) instead of executing a bash command; existing scripts must wrap commands in the prompt (e.g., `check="run npm test — ACK if pass, NACK otherwise"`).

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track Tiny agentic loop with Docker sandbox

Get notified when new releases ship.

About Tiny agentic loop with Docker sandbox

All releases →