Toolbay, the AI marketplace for agents, skills and prompts

Loading…

Here are the bugs in the agent stack you’re using.

With file names and line numbers. We forked Garry Tan’s gstack, hardened it, and measured both.

less context53 paired skills

83%

bugs it missed4 ties, 1 both wrong

8/13

engine selftestsall green

145

20 free skills, no account

$npx toolbay add pr-reviewer

The Stack itself is v0.2.0 and not public yet. Read the full measurement, or run the line above and have a working Claude Code skill in about ten seconds.

Open them yourself

Every one of these is either read straight out of a gstack install or reproduced by a named scenario whose raw log is written to disk.

A safety guard that allows everything when it is missing

gstack/investigate/SKILL.md:26

[ -x "$S" ] && bash "$S" || exit 0

A failing test suite that reports success

gstack/ship/sections/tests.md:170

bin/test-lane 2>&1 | tee /tmp/ship_tests.txt &

An expiry that cannot be parsed means never expires

gstack/ios-qa/daemon/src/allowlist.ts:56

if (Number.isFinite(exp) && exp < now) continue;

The context number is the same kind of claim. A four-skill session (investigate → review → ship → context-restore) injects 38.4 KB here against 290.4 KB upstream, and 134 KB of that is a preamble an earlier skill in the same session already sent. Bytes are measured; tokens are an estimate and are labelled as one everywhere they appear.

The stack can buy the skill it is missing

Every agent stack dead-ends the same way: no skill matches, so it answers directly and you find out later. The toolbay skill is the one thing a fork can add that upstream structurally cannot — somewhere to go. Blocked mid-task, it searches this marketplace. Finished something reusable, it can list it for you. Results are always priced, an empty result stays empty, and the LOCK mode refuses the whole thing before a packet is sent.