npm install -g opencode-goal-mode && opencode-goal-mode --global

Every coding agent will happily announce "✅ Done!" over a half-finished feature and a red test suite. Goal Mode ends that. It moves the discipline out of the prompt — where a confident model just talks past it — and into the harness, where it's enforced in code. The agent's "Goal Completed" is intercepted and rewritten to "Goal Not Completed" unless every required review gate has a fresh pass. Dangerous shell commands are blocked before they ever execute.

It's the difference between asking an agent to be careful and making it.

Watch it refuse to lie

The agent tries to declare victory early. The guard catches it and hands back the truth:

- Goal Completed
+ Goal Not Completed
+
+ Goal Guard blocked completion: required review gates are missing or stale
+ (goal-security-reviewer, goal-final-auditor). State: active=true; dirty=true;
+ reviewCycles=1; missingGates=goal-security-reviewer goal-final-auditor

The agent reaches for something irreversible. The guard kills it before it runs:

$ rm -rf build
✕ Goal Guard blocked a destructive or high-risk bash command
  (rm with recursive force deletion). Use a safer, reversible command
  or ask the user to confirm.

_{↑ While a goal runs, Goal Mode takes over the TUI sidebar with a live, evidence-aware todo list: the goal title, gate progress, and a row per acceptance criterion and missing reviewer — each ticking off as the work gets verified.}

Why you'll want it

🔒 "Done" actually means done. Completion is gated on real review verdicts, not vibes. The model can't answer Goal Completed until every required reviewer returns Verdict: PASS after the last edit — and the claimed Review cycles: N has to match the counter the guard kept.

🤖 The reviews run themselves. When the agent stops with work outstanding, the guard launches the reviewer subagents itself — security, diff, verification, and more — reads their verdicts, and loops fix → review until they pass. You never rely on the model to remember to check its own work.

♻️ One edit reopens the gates. Approvals are stamped with a monotonic sequence, so any change after a review instantly goes stale and forces the relevant reviews to re-run. There's no sneaking a "fix" in after the green light.

🧠 It knows which experts to call. Touch auth and the security reviewer becomes mandatory. Touch a migration and the data reviewer joins. API, performance, tests, UX, ops, docs, quality — the right specialist gates are required automatically from your goal and your diff.

🛡️ Your repo survives. A real shell tokenizer — not a brittle regex — blocks the destructive stuff even when it's disguised: $(rm -rf …), bash -c "…", /bin/rm, busybox rm -rf, git reset --hard, curl | sh. Harmless look-alikes like git checkout -b sail right through.

🚀 It doesn't quit on you. An idle-but-unfinished goal gets automatically pushed forward — told exactly what's left — until it's genuinely complete, with hard caps and a no-progress breaker so it can never spin.

The numbers

Tested on 704 real-world commands from tldr-pages (common/linux/osx) — commands written by hundreds of contributors who've never seen this guard:

And it's effectively free: ~1µs per command, hundreds of thousands of classifications a second. Run it yourself with npm run bench.

How it stacks up

Full side-by-side with citations: research/goal-mode-comparison.md.

Install

One command. Needs Node 20.11+ and OpenCode. macOS and Linux:

npm install -g opencode-goal-mode && opencode-goal-mode --global

Then restart OpenCode. That's it. The installer drops the Goal agent, its reviewer subagents, slash commands, and the guard plugin into ~/.config/opencode, and registers the live sidebar. In the agent picker you'll see just goal — the reviewers are subagents it drives for you. It's idempotent (re-run to upgrade), never touches files you've edited, and --uninstall removes exactly what it added. Goal Mode uses whatever model and provider OpenCode is already set up with.

# Preview first — no writes on --dry-run
opencode-goal-mode --global --dry-run

# One-off with npx (no global package needed)
npx opencode-goal-mode --global

# Into a single project (writes ./.opencode)
npx opencode-goal-mode

# Remove everything it installed
opencode-goal-mode --global --uninstall

# From source
git clone https://github.com/devinoldenburg/opencode-goal-mode
cd opencode-goal-mode && npm ci && npm run install:global

Quick start

# After installing + restarting OpenCode, confirm it loaded — you'll see ONLY
# `goal (primary)`, with every specialist as a (subagent):
opencode agent list | grep goal

Then, in OpenCode, just give it a goal:

/goal add rate limiting to the login endpoint and prove it works

It writes a contract, delegates research to subagents, implements, and verifies — then stops and lets the guard run the reviews. It won't say Goal Completed until they pass. Want to feel the seatbelt? Ask it to rm -rf build mid-session and watch the guard slap it down.

See ARCHITECTURE.md for the full design and research/ for the platform reference, comparison, and threat model.

Configure it (or don't)

Goal Mode works great with zero configuration. When you want to tune it, set options in opencode.json or GOAL_GUARD_* environment variables:

{
  "plugin": [
    ["./plugins/goal-guard.js", { "blockDestructive": true, "contextualGates": true }]
  ]
}

Slash commands: /goal, /goal-contract, /goal-review, /goal-evidence-map, /goal-status, /goal-repair, /goal-final.

Tools the model can call: goal_contract, goal_evidence, goal_evidence_map, goal_reviewer_memory, goal_status, goal_reset.

Troubleshooting

• opencode agent list doesn't show goal? The agents didn't land where OpenCode reads them — re-run opencode-goal-mode --global and restart OpenCode.
• No sidebar todo section? TUI plugins load from tui.json, not plugins/. Confirm ~/.config/opencode/tui.json lists opencode-goal-mode, then fully restart OpenCode. The sidebar is experimental and only shows inside a Goal session with a goal set; enforcement works regardless of the sidebar.
• Reviews didn't kick off on their own? The auto-review fires on session idle. A few free models stall mid-turn without ever going idle, so on those it may not run live — /goal-review and /goal-final run a cycle on demand, and the completion guard still blocks an unearned Goal Completed either way.
• A safe command got blocked? Run node benchmarks/external.mjs --json to see how the analyzer reads it, set blockDestructive: false for that project, and please open an issue.

Good to know

• Requirements: Node 20.11+, OpenCode configured to load local agents/commands/ plugins (tested against @opencode-ai/plugin 1.17.6, compatible with the 1.15+ hook surface), and a working provider/model. Agents inherit your OpenCode default model.
• Safety: The installer copies only agents/*.md, commands/*.md, and plugins/ — never auth files, tokens, or provider config. The guard is a guardrail, not a sandbox, and fails open on input it can't parse; see SECURITY.md for the threat model and a private reporting channel.

Contributing

PRs welcome — CONTRIBUTING.md has the dev loop and release process, and CHANGELOG.md has the full history. Releases are automated and version-synced: one pushed vX.Y.Z tag runs the CI gate, publishes to npm, and creates the matching GitHub Release.

License

MIT · built for OpenCode.