Runtime watch: what DriftLoom tested on 2026-03-15

DriftLoom kept testing skills on March 15, 2026, and the latest wave is the kind of thing this site should be good at: showing what actually happened, not just whether a repo looks tidy from ten feet away.

Today’s tally so far: 40 baseline safety passes, 30 follow-on functionality passes, and 10 failures worth reading, not hand-waving.

What stood out

deai-image (swaylq--deai-image)

This one failed functionality-v2 and currently reads as first observed failure. It passed 9 of 10 checks, so this is not a total wipeout — it is a concrete break worth taking seriously. Failure class: runtime_failed. The first visible tripwire was python help (python-help-ok:). For a newcomer, the takeaway is simple: don’t read this as “mysteriously risky”; read it as “the testing engine found a specific place the skill still falls over.”

deai-image (swaylq--deai-image)

This one passed baseline-v3 and currently reads as passing without failed checks at 8/8. That gives it a solid baseline safety receipt, even if deeper functionality evidence may still be coming later. For a newcomer, this makes the skill easier to browse with confidence, not because it is magically “safe,” but because there is fresh proof on the page.

tmx-cli (lars147--tmx-cli)

This one passed functionality-v2 and currently reads as passing without failed checks at 8/8. That means it cleared the baseline safety lane first and then survived the deeper follow-on checks. For a newcomer, this makes the skill easier to browse with confidence, not because it is magically “safe,” but because there is fresh proof on the page.

tmx-cli (lars147--tmx-cli)

This one passed baseline-v3 and currently reads as passing without failed checks at 8/8. That gives it a solid baseline safety receipt, even if deeper functionality evidence may still be coming later. For a newcomer, this makes the skill easier to browse with confidence, not because it is magically “safe,” but because there is fresh proof on the page.

clawdbot-skill-update (pasogott--clawdbot-skill-update)

This one passed functionality-v2 and currently reads as passing without failed checks at 8/8. That means it cleared the baseline safety lane first and then survived the deeper follow-on checks. For a newcomer, this makes the skill easier to browse with confidence, not because it is magically “safe,” but because there is fresh proof on the page.

clawdbot-skill-update (pasogott--clawdbot-skill-update)

This one passed baseline-v3 and currently reads as passing without failed checks at 8/8. That gives it a solid baseline safety receipt, even if deeper functionality evidence may still be coming later. For a newcomer, this makes the skill easier to browse with confidence, not because it is magically “safe,” but because there is fresh proof on the page.

system-health-check (satoshistackalotto--system-health-check)

This one passed functionality-v2 and currently reads as passing without failed checks at 6/6. That means it cleared the baseline safety lane first and then survived the deeper follow-on checks. For a newcomer, this makes the skill easier to browse with confidence, not because it is magically “safe,” but because there is fresh proof on the page.

usdc (swairshah--usdc)

This one passed baseline-v3 and currently reads as passing without failed checks at 8/8. That gives it a solid baseline safety receipt, even if deeper functionality evidence may still be coming later. For a newcomer, this makes the skill easier to browse with confidence, not because it is magically “safe,” but because there is fresh proof on the page.

What this means if you are browsing casually

Prefer skills with fresh passes, richer follow-on evidence, and clean publisher quality summaries. Treat repeated failures and regressions as useful warning lights, not as drama. The whole point is to make the site readable enough for a newcomer while keeping the raw receipts visible for anyone who wants to inspect the technical details themselves.