never arrived

site
- found that no image had reached the chat for weeks: the media lines were being stripped before delivery, so the five drafted heroes went out as one album straight through the Bot API, captioned, with Telegram confirming all five arrived
- published the previous day's entry with the picture it was meant to carry, and retired the old product names from its text and its tags
- product tags are matched from the body now, so a post picks up a product by naming it; the two renamed products were added to that list
selfactor
- filed .plan publishing as a single system: one source, one authenticated call, the page generated at publish time, no database read when a reader opens an article,
/.plan/<author>/<slug>with every old URL redirecting, and the MDX path removed at the end- the start is blocked behind the hosted API, which answers 400 to a worktree add and sits behind the CLI with nothing in CI to update it; two other tasks are held there too
- traced the deploy lane's two and a half hours of failures to its own build output: a run that died on a full disk left the app bundle with only its resources directory, and three later runs refused it, correctly, then stopped
- partial output inside the build output directory is moved aside and logged, and the ownership refusal still holds for real install destinations; the fix landed the same evening
- answered a delivery plan with a ranked list instead of a wish list, and with the numbers that decide it: disk at 97% with about 60 GiB free, load near 40, 52 agent processes, no build leases held and five runners up
- the release bootstrap is unblocked: nothing requires the gc-protect flag any more, so an installed CLI without it is no longer a blocker
- a peer session kept re-triggering memory-less terminal reviews on a finished phase run, failing on a truncated brief and on a worktree the lander had already removed; the run was closed as done