v0.39.0
Added
get_buildreturns the card of a single build over MCP. The listing was there, the card was
not, so an agent after a manifest name, a vendor or a build comment had to drop to the CLI. The
tool is addressed by build version, and an id works too.
(#14)apps createandapps ensureverify the apply themselves, through--verifyand the
verifyparameter over MCP. The check used to live only indeploy,apps applyand
verify-deploy, so a freshly raised stand needed a second call. The report arrives in the
verifyfield. (#5)
Changed
- The summary under a build listing judges by gaps in the numbering, not by a threshold of
thirty. The platform hands out the numbers of one base version in sequence, so a missing
number is a build it has already removed.ASSEMBLY_STORE_LIMITwent with the threshold.
(#13) buildandinspectrefuse the connection options.--env-file,--base-urland the
credentials used to be dropped without a word, which made the call look like a build tied to a
stand. A behaviour change: it is a refusal with exit code 1 now.
(#7)- The build listing is described the way the platform really works. We thought it capped
builds per project and pushed the old ones out; in fact it deletes the builds nobody uses. The
listing summary and the Console API pages were rewritten.
(#6) --waitends with a check instead of taking the card on trust. A failed apply rolls back to
the previous build and the application comes up running all the same, which the card does not
show. A behaviour change: a failed check exits 1, and--no-verifybrings the plain wait back.
(#5)
Fixed
- The correction about automatic build deletion reached the remaining pages and hints. The CLI
requirements, the MCP server pages, thelist_buildshint and the code comments still told of a
store limit. There is one model now. (#9) builds getworks at last: a build card is addressed by its version. Our pages claimed the
method takes a UUID only, so the command turned the version into an id and got a 404 back. Live
calls on two installations proved otherwise, and both forms are accepted now.
(#8)
Documentation
- The shared
docsguardpackage took over the process-encoding check. It was written here,
though the engine and the bridge start processes the same way and face the same silent failure.
What stays here is the list of folders; the pin goes up tov0.4.0.
(#19) - The shared guard is installed by tag, and raising the pin is a change of its own. On
@main
adocsguardcommit landed in a run here in the middle of unrelated work, and a red run caused
by no commit of ours is one nobody reads.tests/test_workflows.pyfails when the workflows
name different tags. (#18) - A process read as text has to name its encoding, and a test now says so. A new script did
not, and the failure was silent: the Russian page names came back as replacement characters
while the exit code kept reporting success. The convention is written down inCLAUDE.md.
(#17) - The claim mechanics moved into the shared guard. One fact told in several documents at once
is not an elemctl problem: the engine and the bridge keep their documentation the same way.
Claim,claim_textsandclaim_problemscome fromdocsguard.
(#16) python scripts/rebuild-docs.pyrebuilds every generated page in one run. Two generators
write four pages, and each was a thing to remember on its own; the mirrors went first. The
command answers with an exit code, so a failing generator no longer hides the other.
(#15)- The guard catches one statement told in two different ways. A fact about the platform lives
in the specification, on the Console API and MCP pages, in the README and in the code comments
at once, but it gets corrected in one of them. Such facts are listed asCLAIMSin
scripts/check_docs.py. (#12) - The pull request link and the mirror rebuild are one step. A link can only be written once
the pull request exists, so it arrives in a commit of its own, and the mirrors kept being
forgotten.python scripts/changelog-link.py <number>does both halves.
(#11)
What's Changed
- Создание приложения проверяет, что сборка правда применилась by @keyfire in #5
- Перечень сборок: платформа удаляет неиспользуемые, а не вытесняет старые by @keyfire in #6
- A local-only command refuses the connection options by @keyfire in #7
- A build card is addressed by its version, not by its id by @keyfire in #8
- One model of build housekeeping across the pages by @keyfire in #9
- Wrap the new changelog entries like the rest of the file by @keyfire in #10
Full Changelog: v0.38.0...v0.39.0