flaky e2e: is-action (production) 'target action re-runs the loader; decoy action does not' — 2x in one day on unrelated PRs #675

e2e/is-action.test.ts:28 (production project) failed all 3 in-run attempts twice on 2026-07-04, both times on PRs that could not affect it, and passed on plain job rerun both times:

• PR docs(skills): measured perf guidance — async include sizing, cache() economics, typecheck scaling, CF timing caveat #668 (markdown-only skills change): failed run 28711670247, passed on rerun.
• PR fix(ppr): consumption-lane rule — handler ctx.use(loader) bakes during shell capture #674 (PPR capture-gated change, no action-flow involvement): failed run 28716490345 job 85158750998, passed on rerun.

Each failed attempt takes ~10.7-11.3s (suspiciously at/near a 10s expect timeout). Dev project has not shown it. Suggested next step for whoever picks it up: reproduce with repeat-each on the production preview server and inspect what the 10s wait is on (loader re-run signal after the target action?); if timing-bound, widen or restructure the signal rather than the timeout.

Cost: each occurrence burns a full CI rerun (~12 min) and erodes trust in the production shard.