Which agents work with it?

Any agent that reads skills from a directory. Claude Code, Codex and Gemini CLI read it natively; Cursor and VS Code connect over MCP.

The voice runs on your own machine, so there is nothing to meter. The first run downloads it, about 340 MB.

Yes. The skill is MIT-licensed and the authoring format is public.

Create

gstack 1.99.4.0: /ship keeps its second-model review when OpenAI refuses it

9 hours ago

@gstack Subscribe

gstack 1.99.4.0: when OpenAI refuses /ship's second-model review under its cyber-security policy, Capy's /ship retries it on GPT-6 Astra, then Gemini 4 Carbon. Reviewers now state their blocker count, gbrain setups under GBRAIN_HOME sync to the right brain, and three /ship fixes land. Changelog 1.99.4.0: https://github.com/garrytan/gstack/blob/main/CHANGELOG.md#19940---2026-11-14 Commits: https://github.com/garrytan/gstack/commits/main gstack: https://github.com/garrytan/gstack

Ask about this presentation

Show transcript

When OpenAI refused /ship's outside review as a cyber-security risk, that second read was lost. On Capy, /ship now asks GPT-6 Astra again for a read-only correctness review, and then Gemini 4 Carbon.
Slide text

gstack’s /ship keeps its second-model review when OpenAI refuses it

outside voice: OpenAI declined · cyber_policy

RETRY → GPT-6 Astra · read-only correctness review

RETRY → Gemini 4 Carbon · brief unchanged

Refusals Limits Blockers gbrain Fixes
garrytan/gstack · CHANGELOG.md
If all three models refuse, the native review still runs, and the pull request lists the missing read as a coverage gap. The Codex CLI and API runners name the refusal but do not retry.
Slide text

If every model refuses, the pull request says the second read is missing

Still runs: the native review

PR body: lists the gap as missing coverage, never as a pass

No retry yet: the Codex CLI and API runners report the new reason only

Refusals Limits Blockers gbrain Fixes
garrytan/gstack · CHANGELOG.md
Every blocking reviewer in /ship now ends with one line counting its blocking findings. If gstack parsed fewer blockers, it adds a critical row for each missing one, so a blocker the parser misses still blocks.
Slide text

/ship’s reviewers now state how many findings block, and gstack checks the count

REVIEW_VERDICT: blocking=3

findings parsed: 2 blocking

→ CRITICAL row added for the missing blocker

Refusals Limits Blockers gbrain Fixes
garrytan/gstack · CHANGELOG.md
With two brains on one machine, gstack read the config in the home folder and ignored GBRAIN_HOME, so the source wireup could sync to the wrong brain. The wireup, skill start and health checks now follow GBRAIN_HOME.
Slide text

gstack now finds gbrain’s config under GBRAIN_HOME, so syncs reach the right brain

GBRAIN_HOME=/srv/brain

~/.gbrain/config.json

→ /srv/brain/.gbrain/config.json

Source wireup, skill start and /health all follow it. Nothing changes when GBRAIN_HOME is unset.

Refusals Limits Blockers gbrain Fixes
garrytan/gstack · CHANGELOG.md
Tests in hidden folders like the GitHub scripts folder now run instead of failing. The outside review renders again for pull requests over two megabytes, and a results manifest you exclude by name leaves the review.
Slide text

/ship stops failing on hidden test folders and oversized pull requests

Hidden folders: tests under .github/scripts/ now run on Bun 1.4 instead of failing the lane

Over 2 MB: the second-model review renders on its own again

Data files: a manifest.json named by file name in review_exclude_paths leaves the review diff

Refusals Limits Blockers gbrain Fixes
garrytan/gstack · CHANGELOG.md
If you run /ship on Capy, or keep a brain under GBRAIN_HOME, I would take this update, though the model retry needs a Capy machine. The gstack changelog has every detail, including a check for misdirected syncs.
Slide text

Running /ship on Capy, or gbrain under GBRAIN_HOME? Take this one.

gstack CHANGELOG.md

Each change in full, with a check for syncs that went to the wrong brain.

Refusals Limits Blockers gbrain Fixes
garrytan/gstack · CHANGELOG.md