Repository navigation
4 pr 5 release 0.3.1 - #42
Merged
Merged
Conversation
.github/ci-python-inputs.yml lists the repository paths that can change the outcome of the hosted Python job: devcapsule-src, the console, the workflows, and the few files legacy tests still read from the repository, each with its reader named. tests/conftest.py enforces the list on the suite. An audit hook on the interpreter's open, os.listdir and os.scandir events fails a test that opens or lists a repository path outside it, at the access, and the session fails when a breach was recorded. A test does not count when the hosted job does not run it or passes whatever it does: a heavier suite's marker, or an expected-to-fail mark with no condition and no strict. The workflow still routes by .github/ci-paths.yml in this commit. The next one makes it read this list, once the guard has run on the hosted runner, where the tests marked expected to fail inside a capsule run for real. For the bug "Markdown-only changes run the whole Python unit suite". Co-Authored-By: Claude <noreply@anthropic.com>
The workflow reads .github/ci-python-inputs.yml with its own filter step and the default quantifier: one changed file under one entry runs the unit suite, and any other change skips it, a new directory included. .github/ci-paths.yml keeps the docs route alone. The guard of the previous commit makes the skip sound: no counted test can read a repository path the list leaves out. tests/test_ci_reporting.py was a legacy file and its filter test encoded the old default, so it is refurbished whole under Unit Tests rule 5: ci_gate.py and diff_coverage_annotations.py get their mirrors under tests/unit/scripts and declare their public surface; the route script, both filter files and the real diff-cover run become integration tests in tests/integration/test_ci_workflow.py. docs/** stays an input: one legacy scan reads all of docs/ and counts on the hosted runner, where the old filter skipped it. It leaves the list when that scan is retired. Records: the bug is re-rated major by the owner, targeted at 0.3.1 and in fixing; the blocking record gains the finding that one expected-to-fail unit test downloads 1.5 GB of vendor artifacts inside the owner's capsule on every run. Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Claude <noreply@anthropic.com>
…e-guard Route the Python job by a list of inputs the unit suite is held to
…source PR #18 gave eight production signatures twenty keyword parameters that no production caller passes, so that no refurbished test would patch a devcapsule attribute. Every new test passed the parameter, so the default, the branch a release ships, was run by none of them: of six faults planted in those defaults, the unit suite caught none, where the retired legacy files had caught five. Runtime: entrypoint.run, console_child, plan_filesystem, prepare_filesystem and foreground_command take what they took before PR #18. run's body is the one 0.3.0 shipped, and with it goes the conditional forwarding of caller_uid and runtime_root that kept two legacy tests' stand-in for prepare_filesystem working. console_child also loses launcher_source, added by PR #16 in the same style. The one addition is a name for a fact: filesystem.RUNTIME_ROOT, /tmp. Commands: VersionCommand and CheckoutDecisionsCommand no longer override invoke to carry test inputs through the framework's group-context argument; the private _DecisionsInvocation is gone, and run_console_decisions takes its arguments alone. Tests: the mirror files set each host input where the code reads it. The caller's uid, ownership changes and the readiness connection go through os.geteuid, os.chown and socket.create_connection; the image's fixed locations through the public constants CONSOLE_PYTHON, LAUNCHER_CONSOLE_SOURCE, RUNTIME_ROOT and RUNTIME_PLAN_PATH; for run alone, the supervisor, the display and the log setup are stand-ins. identity.py gets its mirror file. The integration test of the wrapper's help builds a base without a console interpreter again. images/contract.py no longer calls the runtime's check at start a backstop for the recipe: since PR #18 removed the per-start dependency probe, that check finds only an absent interpreter. Evidence: 26 faults planted one at a time in a scratch copy and run against the touched unit files; 24 caught there, the five lost by PR #18 among them; one caught by the integration file only, the decisions command dropping its arguments; one caught by neither, as before PR #18, a built executable that never finds its stamped record. Left alone: current_build_info keeps its three parameters, because the stamped record lies inside the installed package where a test cannot build it; console_decisions_command keeps the four it had before PR #18. Co-Authored-By: Claude <noreply@anthropic.com>
The maintenance status records the findings of the assessment of PR #18, what this slice delivers, the four judgments taken for the owner, the planted-fault evidence, the validation counts in the owner's capsule and the proposals not implemented. It also records that PR #25 is merged. The brief for the Codex and GPT-6 Astra pair is committed under the workstream's evidence. It asks for the hosted check's result and names the capsule's 17 standing unit failures, so that neither is read past. Co-Authored-By: Claude <noreply@anthropic.com>
Correct the console label-gate claim and keep process-wide patches inside test actions. Assert environment timing, graphics defaults, the runtime root, and package-resource build identity. Keep the decisions mirror focused on one unit and isolate its fake interpreter from Python startup settings. Evidence: 81 focused tests, 10 integration tests, and 98 legacy tests pass. Typecheck passes over 336 files. Thirteen planted production faults and one fixture fault are detected. Co-authored-by: GPT-6 Astra Codex <noreply@openai.com>
Codex + GPT-6 Astra review of the removal of PR #18's test-only parameters, accepted without a round. It corrects the label-gate sentence in images/contract.py, which claimed more than the launcher enforces; ends each process-wide test patch with the action it serves; reads the session environment when the supervisor is made and asserts the graphics defaults; asserts the runtime directory under /tmp and not under the constant; keeps the console tools mirror to its own unit; isolates the fake interpreter from Python startup settings; and adds a test of the default build-record lookup, which catches the one planted fault #26 had left uncaught.
PR #27, the Codex and GPT-6 Astra review, was accepted at 1b5d3fc without a round and merged at 60ad2c5. The status records what it changed, and corrects a judgment of mine it disproved: the default build-record lookup can be tested, by building a package directory and replacing files at its lookup site. Validation on the merged branch in the owner's capsule: the unit session gives the 17 standing failures of the blocking record and 2914 passed; type check, CLI smoke, executable build and smoke, documentation contract and console session pass. Hosted run of PR #26 at the merge: 2950 passed. 25 of the 26 planted faults are caught by the unit files, the last by the integration file. The pair review document gains the loop's row. Co-Authored-By: Claude <noreply@anthropic.com>
…ched test patches Owner direction of 2026-11-11: a value an inner module would be patched to supply is a parameter at construction; only the top layer maps the host; a unit test constructs ordinary values; classes bind the values that are used again and again. Product code: - identity: CapsuleUser(identity, caller_uid) replaces foreground_command; its owner_changes decides which new directories change owner. - filesystem: create_directories returns what it created; prepare_filesystem(plan, user) performs the owner changes. RUNTIME_ROOT is removed. - console: ConsoleInstallation(python, launcher_source) with package_path and child replaces console_package_path and console_child; probe_address is public. - session, new: Session(plan, job, user, console) decides the filesystem plan, the surface's launch, the variables to set, the display, the profile lock, the log settings and the children. - entrypoint: run reads the effective uid and the environment once, builds the session and performs its answers. Same signature. - console_tools: DecisionTools(in_capsule, console, runtime_plan_path) replaces two functions with four optional parameters; the command reads the environment. - build_info: distribution_build_info and embedded_build_info take every location as a value; current_build_info and is_built_distribution take none. build_info_text is the text `version` prints. Tests: the mirror files of these units and three integration files hold no monkeypatch use, from 26. The readiness probe meets a real loopback socket. run and the two command classes are exercised by integration tests that start the real programs in an environment the test builds: a headless and an interactive session with a stand-in IDE, `version`, and `project checkout decisions`. Rule 5: tests/test_runtime_entrypoint.py pinned the old signatures under patches, so the whole file is refurbished into the mirrors of contract, filesystem, identity, session, entrypoint, graphics and components/jetbrains, and deleted. graphics and jetbrains declare __all__. tests/test_contained_display.py, legacy, now defines the three helpers it imported from that file; none of its test functions changed. conftest no longer patches console.LAUNCHER_CONSOLE_SOURCE, which no unit test reads through an entry layer any more. Convention check: 102 legacy files, 1350 legacy tests, 0 breaches. Evidence: of 43 planted faults, the mirror files catch 26, legacy unit files 4, the integration files 8, and 5 are caught by nothing: the probe's timeout (two faults), `checkout decisions` dropping its arguments, uid and gid swapped in the one chown call, and the stamped record looked for outside the package. Of the 26 faults of 2026-11-11, the unit suite now catches 21, where the patched tests caught 25. Validation in the capsule: unit session 17 failed, the standing ones of the blocking record and no other, 2959 passed; type check, smoke, PEX build and smoke, docs contract pass. Co-Authored-By: Claude <noreply@anthropic.com>
…patches The maintenance status gains the section of 2026-11-12: the owner's direction, what each unit became, the refurbishment of the legacy file that rule 5 required, the planted-fault account with its losses, the validation in the capsule, four judgments and three proposals. The section of 2026-11-11 is marked superseded in part. The rewrite has not been reviewed by the partner. Co-Authored-By: Claude <noreply@anthropic.com>
…ew of PR #26's rewrite Add the section Host Inputs to WORKFLOW-LOCAL.md, on the owner's word of 2026-11-12: only an entry layer reads the host, inner modules take values in required arguments, frozen dataclasses bind shared values, decisions return data, new and changed unit tests patch nothing, entry layers are tested as programs, and the pull request states what the hosted gate no longer catches. - WORKFLOW-LOCAL-humane.md: the companion paragraph, with the rewrite of PR #26 as the worked example and its cost in planted faults. - engineering-docs/development/unit-tests.md: the convention summary names the rule and test_session.py as the model. - engineering-docs/development/review-partner-loop.md: the brief template gains the reviewer's duty under rule 7. - evidence/2026-11-12-host-inputs-as-values/review-brief.md: the brief for the Codex + GPT-6 Astra pair, PR26.2. Evidence: nox -s docs-contract passes, 249 pages. Co-Authored-By: Claude <noreply@anthropic.com>
Set launch variables only after writing launch files, as in 0.3.0. Use independent version and runtime-plan fixtures. Cover real ownership and decision argument forwarding, and state integration prerequisites. Twelve planted faults produced ten detected failures and two timeout gaps. Validation: 341 files typechecked; 157 mirror, 25 integration, and 69 legacy checks passed. The full unit run passed 2961 tests with the 17 known mounted-settings failures and no additional failures. Co-authored-by: GPT-6 Astra Codex <noreply@openai.com>
… of #26 (PR26.2) Codex + GPT-6 Astra review of the rewrite of PR #26, accepted without a round. Launch variables are set after the launch files are written, as in 0.3.0. The version integration tests assert an independent record from a private copy of the package. The decisions integration test proves that the arguments reach the console. A new integration test proves the uid and gid of the real owner change. A contract fixture no longer inherits PyCharm's paths. Two value boundaries gained mirror tests.
Maintenance status: a new section with the rule, what PR #28 changed, the one remaining difference from 0.3.0, the deviations from the rule in the merged code, and the seven places where the reviewer found the rule open to two readings, for the owner's ruling. Loop record: the 2026-11-12 row. Evidence: nox -s tests on 83d238b in the capsule, 17 failed (the standing ones), 2961 passed; convention check 102 legacy files, 1350 legacy tests, 53 conforming, 0 breaches. Co-Authored-By: Claude <noreply@anthropic.com>
Written by devcapsule0 workflow publish, coordination 1ff67b6. Co-Authored-By: Claude <noreply@anthropic.com>
…command line Unit Tests rules 8 and 9, on the owner's word of 2026-11-12, each held by machine. - scripts/validate_unittests_follow_conventions.py: a fourth check. A test file that is not legacy is refused when it imports devcapsule.cli or devcapsule.__main__, directly or through a module under tests; so is a conftest.py. tests/command_line_debt.txt lists the eight mirror files that do today; the check refuses a stale line and an added line. - tests/conftest.py: NetworkGuard, on the interpreter's audit hook. While a unit test runs, a connect or send to another machine and a lookup of anything but this machine fail the test before anything is sent, and the session fails with a list of the attempts. Loopback stays open; heavier suites are not held to it. - WORKFLOW-LOCAL.md, its companion and the unit test guide state the rules. Evidence: 27 planted faults, 23 caught by the unit tests, 4 by tests/integration/test_network_guard.py, none uncaught. In the capsule, nox -s typecheck passes over 342 files; nox -s tests ends with the 17 standing failures and 3008 passed in 170 s, where it took 815 s: the one test that downloaded 1.5 GB is now stopped at its name lookup. Co-Authored-By: Claude <noreply@anthropic.com>
…machine Co-Authored-By: Claude <noreply@anthropic.com>
Written by devcapsule0 workflow publish, coordination f4a63ee. Co-Authored-By: Claude <noreply@anthropic.com>
…m values Owner direction of 2026-11-12: read the configuration into memory first, decide second, and let a test of a decision skip the reading. - configuration/derivation.py: CheckoutState is a dataclass of values; load_state does all the reading. The state's methods no longer read the home's directory, the record's existence or the console's requests. - commands/_launch_review.py, new: review_launch(state, flags, ask=...) is the decision phase. It returns a LaunchReview or refuses, and writes nothing. RunFlags and Prepared moved here; _run_flow still exports them. - commands/_run_flow.py: prepare_launch is three phases in order: read_launch_state, review_launch, save_launch. The terminal dialogue stays there and is passed to the review as ask. - tests/unit/devcapsule/commands/test_launch_review.py: 18 tests that build their checkout in memory, with no directory, no home and no patch. No caller and no legacy test changed. Evidence: the unit suite passes as before, 3026 passed with the capsule's 17 standing failures; with the mounted-settings fallback switched off it fails only the fallback's own two tests, before and after. 22 planted faults: 17 caught, 5 named in the maintenance status with the reason for each. Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Claude <noreply@anthropic.com>
…ecisions as functions The owner asked why the decision phase still read the host in five places. Three are closed here, the ones no legacy test calls: - derivation.py: Host, two functions bound by load_state. derive_plan asks it for a local base image; the lookup still goes through the operations adapter when called, so the legacy tests that patch it are unaffected. - _launch_review.py: stage_trust stamps an acceptance with the host's time; a run-once authorization goes to announce, a function the review is given. - _run_flow.py: prepare_launch passes standard error as announce. The fourth, the dialogue, was already the function ask. The fifth is left: the listing and the configuration review ask the environment and the file system, in assess_checkout and Configuration, which five legacy files call. Evidence: 20 mirror tests, three of them new, with no directory, no home and no patch; five planted faults, four caught, the fifth the one line that binds the real clock. The unit suite passes as before, 3028 passed with the capsule's 17 standing failures. Co-Authored-By: Claude <noreply@anthropic.com>
…and what the last waits for Co-Authored-By: Claude <noreply@anthropic.com>
…h the decisions through the Host Owner ruling of 2026-11-12, option 3: the argument is optional until the legacy tests that call these functions are retired. - configuration/bindings.py: HostDirectory, what the host says of a named directory, and host_directory, which asks this machine. - assess_checkout takes host_environment and directory; Configuration, review_configuration and resolve_configuration_bindings take directory. Left out, this process's environment and this machine's file system are asked, as before; five legacy files rely on that. - derivation.py: the Host carries both; the state's listing and derive_plan pass them. Nothing in the decision phase reads the machine any more. Evidence: four new mirror tests with a constructed host, 24 in all. The unit suite with the mounted-settings fallback switched off fails only the fallback's own two tests, as before: 3047 passed. Co-Authored-By: Claude <noreply@anthropic.com>
…e's twenty-six tests tests/test_ide_requirement_override.py is ordered retired whole. These are the tests of it that need no product change: - test_launch_review.py: a forced IDE replaces the required one and keeps the rest; a saved version set; the minimum a departure waives and the one it does not; the required IDE named by another spelling; a schema-1 manifest. 33 tests now. - configuration/test_capabilities.py, new: CapabilityPolicy.departing. - configuration/test_capability_selection.py: a saved IDE that conflicts with a later requirement names the override. - capabilities.py and bindings.py declare __all__. The legacy file stays: its other fourteen tests reach the command line with patched launch and materialization, files, a patched component class and a mirror with a legacy twin. The maintenance status lists the five product changes they wait for. Evidence: nox -s typecheck passes over 345 files; nox -s tests, 3045 passed with the capsule's 17 standing failures; 0 convention breaches. Co-Authored-By: Claude <noreply@anthropic.com>
…st legacy file's retirement stands Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Claude <noreply@anthropic.com>
… keyword arguments Owner direction of 2026-11-12, option 2, beside option 3. - WORKFLOW-LOCAL.md: rule 5 allows one change to a legacy test, keyword arguments added to or removed from a call when the function's signature changes in the same pull request. Host Inputs rule 2 names the two functions that keep a host-reading default until their legacy callers are retired. The companion says why, for both. - scripts/validate_unittests_follow_conventions.py: a changed legacy test is accepted when its two versions differ only in which keyword arguments the calls in its body name. Positional arguments, the values of keywords that stay, every other statement, the decorators and the test's own parameters must be the same; a change of layout alone is still refused. Evidence: 12 new cases in the checker's tests, 4 accepted forms and 8 refused ones; nox -s typecheck passes; 0 breaches in the repository. Co-Authored-By: Claude <noreply@anthropic.com>
…side a capsule The file's home fixture sets DEVCAPSULE_CONFIG_HOME under the test's directory. A home that is only implied, with no settings of its own, reads the launcher's mounted settings inside a capsule; a named home does not. No test function and no product code changed. Evidence: all 33 cases of the file pass in the owner's capsule, where two failed; nox -s tests ends with 15 failed, the standing ones of the other five files, and 3059 passed. Co-Authored-By: Claude <noreply@anthropic.com>
A change to a console page could be seen only in a capsule launched from the checkout or after a build. The session installs the console editable with its pinned dependencies and the launcher, makes a token for the run, logs the links to open, the sample presentation among them, and serves this repository, or the project given with -- --project, on port 8787 or -- --port. An edit to a page, a script or a stylesheet shows on a browser reload; an edit to the Python needs the session run again. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The console's presentation page needed a presentation.json written by hand for each recording. A replay now writes it beside the video when the scenario names its script, so a new recording is presented with no further work. - tests/e2e/scenario_presentation.py reads the script's goal, stack, illustrations and numbered steps, joins them to the scenario's actions by the step number that starts each action's reason, and places steps and captions on the video from the count of frames the recorder had taken after each action. A step without captions shows its actions' reasons; an action's "caption" starts a caption that lasts until the next one. - scenario_recorder.run_scenario_replay and todo_demo_controller.replay_mode plan the presentation before anything is recorded, so a script that does not join its scenario stops the run at once, and write the document after the video is encoded. A scenario without a script is recorded as before. - The todo scenario and its script carry the presentation's text: title, summary, what it illustrates, and nineteen captions. With the frame counts of the 2026-11-13 run, the generator reproduces the hand-written sample's steps, texts, captions and times; the sample is now its output, with the step's last screenshot for the seventh step. - running-e2e-scenario-recordings.md, section 5, is the instruction for a scenario's author: the script's form, numbered reasons, captions, and how to look at the result. Checked: nox -s tests (3262 passed, the conventions gate among them), nox -s typecheck, the generator's 39 unit tests at full coverage of the module, and the generated sample walked slide by slide in Chromium. Not checked: a new recording, which needs Docker; the two replay loops' calls are exercised only by one. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…e was merged Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…entations Web console: presentations, a recorded scenario as slides beside its recording, written by the replay itself
Chrome plays a video response through a document of its own that inherits the response's policy, and the raw route's policy forbids that document every fetch, its own player's included. So a .webm opened from Files showed a player that never played. The Files page now plays a video or audio file itself at /files/<path>: breadcrumbs, the player and the download link. The tree route gives each regular file's entry the media type the raw route serves it as, so the page keeps no table of its own. Listing rows link such a file to its /files address, and a link in a record to a file that is not Markdown leads to /files/<path>, where any other file is sent on to the raw route with the address's fragment. The raw route is unchanged. The bug record comes from ws-project-management/coordination with the fix recorded: status fixed, target 0.3.1. Not covered: a symbolic link to a recording. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…kstream's status Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…o-in-files Web console: a video or audio file opened from Files plays in the page
…ing-bugs Records: the mounted-settings bug closed; the .devcapsule refresh checked and waiting for the host
… the raw route A console built before 6ff6bf1 sends a video opened from Files to the raw route, whose policy keeps Chrome's own player from loading it. The extension's button replaces such a tab with a page of the extension that plays the same raw address. Its README says what it is for, how to load it, and that the folder is to be deleted once no console in use predates the fix. Not part of the launcher, the console or any build; no test covers it. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…-player-extension scripts: a temporary Chrome extension that plays a console .webm from the raw route
…ole module names its own interpreter The rotated-consensus folder no longer exists and no tracked file is under it, so its module file and its line in modules.xml go. The web console module excludes devcapsule-webconsole/.venv from its content and names that environment as its interpreter, in the form devcapsule.iml uses for devcapsule-src/.venv. These are the owner's IDE edits, committed as they stood in the checkout. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…module-files IDE project files: the rotated-consensus module removed; the web console module names its own interpreter
…ful behavior Under rootless Docker (SecurityOptions reporting rootless), host bind mounts owned by the host user appear as 0:0 inside the container, while passing --user <host_uid>:<host_gid> runs the container as a subordinate UID that cannot read 0600 /etc/devcapsule/home/settings.toml or write to /home/devcapsule/.cache/pex. Query docker info SecurityOptions at the entry layers (project run and recursive successor) and pass rootless_docker=True plus uid=0, gid=0, user='root' in RuntimePlan only when the daemon reports rootless mode. When rootless_docker is True: - pass --user 0:0, IDE_UID=0, IDE_GID=0, IDE_USER=root, map host-owned Docker sockets to container GID 0, and write /etc/passwd and /etc/group for root with /home/devcapsule - skip redundant host chown/helper steps for sudoers policy files that already map to container 0:0 - skip gosu 0:0 and same-owner chown calls in CapsuleUser, log/display ownership helpers, and entrypoint scripts under --cap-drop ALL When Docker is not rootless, every path and unit-test default stays unchanged. Includes the confirmed/fixed blocking 0.3.1 bug record. Co-authored-by: Antigravity <noreply@google.com>
…ess-docker Launcher: support rootless Docker (0:0 namespace mapping)
… host Owner verified devcapsule-local.pex (v0.3.1-local-linux-x86_64, built from a1a54eb, SHA-256 7921e3bbd6abbe5ade7790d5c9020383d6117aa109608fa2714206246fde5d40) on the Ubuntu 24.04.5 LTS rootless Docker installation. Co-authored-by: Antigravity <noreply@google.com>
Co-authored-by: Antigravity (Gemini 4 Carbon High) <noreply@google.com>
Co-authored-by: Antigravity (Gemini 4 Carbon High) <noreply@google.com>
With website converted from a .gitmodules submodule to a raw gitlink (160000 tree entry without a .gitmodules mapping), git submodule update --init website fails with 'No url found for submodule path website in .gitmodules'. Update tests.yml (docs-build) and website.yml (build and candidate) to resolve HEAD:website directly, fetch that exact commit shallowly over HTTPS from mycodespace-ai/devcapsule-website.git, initialize any nested submodules inside website recursively over HTTPS, and update any other submodules declared in .gitmodules without invoking git submodule update on raw gitlinks. Add an offline integration test in tests/integration/test_ci_workflow.py exercising all three workflow jobs against a raw website gitlink with a nested SSH submodule. Co-authored-by: Antigravity (Gemini 4 Carbon High) <noreply@google.com>
Co-authored-by: Antigravity (Gemini 4 Carbon High) <noreply@google.com>
…link-actions Ws maintenance/website gitlink actions
…ess-docker Records: close rootless-Docker bug after owner verification
Co-authored-by: Antigravity (Gemini 4 Carbon High) <noreply@google.com>
Convert website submodule to raw gitlink
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Test coverage, from CI run
devcapsule