Releases: hangglider5/agent-ground-control
Release list
Agent Ground Control v0.1.0 — Build Week Evaluation
Agent Ground Control v0.1.0 — Build Week evaluation build
Agent Ground Control is a macOS-first menu bar app and CLI that answers one machine-level
question: what did my coding agents leave running?
It attributes local processes, loopback ports, LaunchAgents, browsers, and worktrees to Codex,
OpenClaw, and Hermes Agent runs. A deterministic safety gate previews cleanup, revalidates process
identity, refuses ambiguous or dirty-worktree cases, uses graceful termination, and records an
immutable receipt.
Download
Agent-Ground-Control.dmg— macOS 14 or later, Apple Silicon- SHA-256:
f523cc657af49535b502fdfa5d52628bddfbe18e3ff26bcde02d96f70d80fe9c
Verify after downloading:
shasum -a 256 Agent-Ground-Control.dmgEvaluation-build notice
This hackathon test build is ad-hoc signed. It is not Developer ID signed or notarized. After
verifying the checksum, macOS may require System Settings → Privacy & Security → Open Anyway
on first launch. The README contains the complete installation, Login Items, Demo Lab, CLI, and
uninstall workflow.
The exact package passed the complete warning-as-error release gate. The packaged service
lifecycle also passed an installed SMAppService register/run/unregister smoke during release
qualification. No agent account, API key, runtime LLM, root access, or network connection is
required for the deterministic Demo Lab.
Fast judge path
- Drag the app from the DMG to
/Applicationsand open it. - In Inspector → Settings, enable automatic daemon startup.
- In Demo Lab, choose Start simulated suite.
- Inspect the orphan-browser evidence and dirty-worktree refusal.
- Compare the UI cleanup preview with the shipped
agentctlCLI. - Confirm one eligible cleanup and inspect its immutable receipt.
- Stop Demo Lab and disable automatic daemon startup before uninstalling.
See docs/testing/judge-guide.md for the ten-minute no-rebuild walkthrough.
Source code is licensed under Apache-2.0.
What Codex and GPT-5.6 contributed
Codex with GPT-5.6 helped research upstream agent behavior, refine the product and safety
architecture, implement the native Swift modules test-first with subagents, run independent
reviews, and close integration, privacy, performance, and packaging defects. The shipped runtime
is deliberately classical and deterministic: no LLM API is called by the app.