# Reader FAQ: the `stack apply` changelog entries Short answers about what `stack apply` changed and where the change is recorded, for a reader of the two checked-in changelogs: the repository [CHANGELOG.md](../CHANGELOG.md) and the site copy [packages/site/src/data/CHANGELOG.md](../packages/site/src/data/CHANGELOG.md), which the site renders at `/docs/changelog`. These answers describe the checked-in files, not an installed version or a deployed site; compare `stack apply --help` for the version in use. ## What changed? `stack apply`'s wiring step stopped attempting to write placeholder secrets to the Phantom vault and instead checks declared secret names. The fix is recorded as the first bullet under the first **Bug fixes** heading of **0.2.1 (2026-10-16)**, titled "`stack apply` stops attempting to write placeholder secrets." Per that entry, wiring previously ran `phantom add TODO_STACK_APPLY --rotate` (then without `--rotate`) for every declared secret. Phantom Secrets refuses values in argv and has no `--rotate`, so every call failed and the printed `envelopes` count was always 0; `WireResult.envelopes` is replaced by `present` and `missing`. The wiring step never writes to the vault or fills in missing values: each required missing value needs user action — run `phantom add ` from a trusted terminal and enter the real value — and provider setup may separately store real values. ## Where are the entries found? The same wording appears in both files. In each: - [Unreleased → Changed](../CHANGELOG.md#unreleased) ([site copy](../packages/site/src/data/CHANGELOG.md#unreleased)) — current wiring behavior: after at least one provider succeeds, names are checked for every recipe provider including failed ones; a missing name does not itself fail apply, while provider failures do; `--noWire` skips both checks and stubs. - [0.3.0 (2026-10-31)](../CHANGELOG.md#030-2026-10-31) ([site copy](../packages/site/src/data/CHANGELOG.md#030-2026-10-31)) — the summary under **Changed and fixed** plus the detailed **Changed** entry with readable- and unreadable-vault output examples. - [0.2.2 (2026-10-25)](../CHANGELOG.md#022-2026-10-25) ([site copy](../packages/site/src/data/CHANGELOG.md#022-2026-10-25)) — related `stack apply` fixes under **Bug fixes**: a missing recipe writes nothing, and a failed provider no longer stops later providers. Not the placeholder change itself. - [0.2.1 (2026-10-16)](../CHANGELOG.md#021-2026-10-16) ([site copy](../packages/site/src/data/CHANGELOG.md#021-2026-10-16)) — the fix, first bullet under the first **Bug fixes** heading. The release has a second **Bug fixes** heading under "Earlier 0.2.1 notes (2026-10-12)", so the release anchor, not the subsection name, is the stable link target. - [0.1.0 — 2026-04-19 (superseded by 0.1.1)](../CHANGELOG.md#010--2026-04-19-superseded-by-011) ([site copy](../packages/site/src/data/CHANGELOG.md#010--2026-04-19-superseded-by-011)) — the original `stack apply [recipe-id]` description under **AI recommendation layer**, annotated with the failed placeholder history and a pointer to the 0.2.1 fix. ## Do the two changelogs differ? Not in this checkout: the files are byte-for-byte identical. `CHANGELOG.md` is the source of truth; the site's `sync-content` script overwrites `packages/site/src/data/CHANGELOG.md` with it before site development and builds, so a site-only edit would be overwritten by the next sync. Verify rather than assume, from the repository root: ```sh diff -u CHANGELOG.md packages/site/src/data/CHANGELOG.md ``` No output and exit status 0 mean they match; status 1 prints the differences. ## Which release shipped the fix? No release labelled 0.2.1 reached npm: the changelog says it was tagged but its publish failed, and 0.2.2 supersedes it, shipping everything under both headings. The fix therefore first shipped in 0.2.2; the 0.3.0 and Unreleased entries clarify the behavior. ## What do the entries not establish? - A missing name alone does not fail apply; provider failures do, so a successful apply is not proof that every required secret is present. - The counts describe checked names. With an unreadable vault — or an unknown provider — providers are reported as skipped and the zero counts do not show the vault is empty or complete. - Wiring can create webhook stubs even if the vault is unreadable; the entries describe stubs, not finished handlers. - The files do not record which version is installed or deployed. ## How do I verify these answers? ```sh grep -n -F 'stops attempting to write placeholder secrets.' \ CHANGELOG.md packages/site/src/data/CHANGELOG.md diff -u CHANGELOG.md packages/site/src/data/CHANGELOG.md HOME=$(mktemp -d) bun test packages/site/tests/changelog.test.ts ``` Expect exactly one `grep` hit in each file, inside 0.2.1 → Bug fixes; no `diff` output; and the site tests to pass, including their byte-equality check. For a prose walkthrough see the [reader guide](changelog-apply-reader-guide.md); for the entry table see the [source index](apply-changelog-source-index.md); for version-by-version reading see the [timeline](apply-changelog-reader-timeline.md) and the [glossary](apply-changelog-reader-glossary.md).