- ForgeActionFailure gains `unknown` for an answer that does not confirm
the outcome; the GitHub provider passes it on instead of `failed`, the
panel Remote refuses with `forgePanel/action-unknown`, a viewed mark
stays, and the other actions ask the reader to check the browser.
- Reruns skip workflow runs with a pending check, try every finished run,
and report the refused ones together; the card offers a rerun only when
a failed check's run has finished.
- Review requests resolve users and teams to node ids and call
`requestReviews`, which GitHub Enterprise Server 3.19 and 3.20 have;
bots still go through `requestReviewsByLogin`.
- Viewed marks are kept per change request and head, so a reread at the
same head keeps writes in flight and their locks, and paths named after
object properties read only their own entries.
- The reviewer search never chooses from the listing of an earlier query.
- Documentation states the unknown code, the branch update evidence, and
the `canBeRebased` approximation, and drops stray blank lines.
The forge subsystem page gains the overview and viewed mark types and
contracts, and the generated Cordis, configuration, and event catalogs
follow the new service methods and the maxViewedFiles option.
ChangeRequestSnapshot now carries the reviewers, linked closing issues,
branch update methods, and the viewer's rights to request reviews and
rerun checks; each check names its rerun. Providers may implement viewed
mark reads and writes, failed check reruns, review requests with a
reviewer search, and head-pinned branch updates. The registry fails an
action a provider lacks without polling, and polls a watched change
request after every action the provider answered.
A write to a thread or comment now reads, in the head check, the pull
request that holds it and is refused without a mutation when that is
another pull request. Submitting the pending review keeps the
submission to reconcile, so an unknown outcome blocks a second review.
GraphQL errors count as refusals only when every type names one, so a
GitHub timeout reports an unknown outcome. Pending reviews are found
by viewerDidAuthor among up to 100, without the unused commit field. A
resolved thread stays open while the reader writes in it, unsupported
writes no longer poll, and the docs state the partial two-step write.
The Harness-home exemption matched every name case-insensitively, so a
home `.env` line such as `deepseek_base_url=...` passed the check and
landed in `process.env` under that spelling. The launch snapshot looks
names up exactly on POSIX, so llm-deepseek, web-search-deepseek, and
web-deepseek-api never saw it: the launch succeeded with the official
endpoint, where it used to fail. Proxy names keep matching in any letter
case because proxy clients read both spellings; the two endpoint names
now match only as `DEEPSEEK_BASE_URL` and `DEEPSEEK_SEARCH_BASE_URL`, and
another spelling fails the launch with the exact name to use.
The home-layer refusal case now covers `SSL_CERT_FILE`,
`NODE_TLS_REJECT_UNAUTHORIZED`, and `DSH_HOME`. The web-deepseek-api
README states where `DEEPSEEK_BASE_URL` may come from; the
launch-environment README, the development guide, and the configuration
source Agent Note no longer overstate which file is refused, which
Vitest configurations load the root `.env`, or what a checkout may carry.
Refs #8138
- A jump to a thread centered its card on every later jump to the same file, since the jump target outlived the jump.
The scroll target now carries the card to center and is cleared when the jump ends; unfolding still follows the
latest thread jump.
- In a commit range view, a thread on a file the chosen commits do not change opened on the host instead of choosing
all commits; a thread the chosen commits do not hold now always chooses all commits first.
- A quote of an old-side thread named the head as the commit its lines number. It now says the lines number the file
before the change and names the head only as where the diff ends, and ForgeAnchor documents the same.
- Refresh reads review threads from the host past the panel's cache (`changeRequestReview` takes `fresh`), so a thread
resolved since shows without waiting for `reviewRefreshMs`; `maxCachedReviewBytes` defaults to 16 MiB, at least
one read under forge-github's default `maxOutputBytes`.
- Picking a thread from a list that covers the files moves the focus to the thread's card.
- `ForgeReviewData.viewer`, its GraphQL field, `ThreadActions.viewer`, and `ThreadReadState.unsupported` had no
consumer and are removed.
- A quoted outdated thread is named by the lines it was written on; a missing pull request or repository makes a
review read `not-found`; a truncated read that kept no thread still offers the thread list.
`loadLayeredEnv` refused `DEEPSEEK_BASE_URL` and
`DEEPSEEK_SEARCH_BASE_URL` from every discovered `.env`, including the
user's own `$DSH_HOME/.env`, so a machine-level endpoint written there
made both the CLI and Desktop fail at startup.
The refusal exists for the invoking directory's file: it arrives with a
clone, and a repository must not choose where requests carrying the
user's DeepSeek credential go. The home file is the user's own, and the
profile patches in the same directory can already set these endpoints as
plugin `baseURL` fields, so refusing it there protected nothing.
`HOME_LAYER_PROXY_NAMES` becomes `HOME_LAYER_ROUTE_NAMES` and adds the
two endpoint names; a project `.env` that sets either still fails the
launch, and its message now names the home file as the second way out.
`DSH_HOME` stays bootstrap-only, and the CA and TLS names stay refused
everywhere.
Tests cover the project layer refusing both names, the home layer
accepting both below an exported value, and the project-layer message.
The app-boot, launch-environment, llm-deepseek, and web-search-deepseek
READMEs, the development guide, and the two Agent Notes that stated the
old rule now state this one.
Refs #8138
- Return the early answer from scanCommand without awaiting a successful
outcome: the SSH provider rejects done when termination cancels a
launch it has not yet acknowledged, which turned a retained dirty
checkout into a failed cleanup.
- Test that a never-initialized submodule is removed and that Git
refuses a committed embedded repository, which inspection leaves to
Git by design.
- Record the git-status-only trade-off and its sparse-checkout rationale
in the worktree Agent Note, and align the README, upgrade guide,
maxOutputBytes description, and comments with what is streamed.
Refs #8151
The forge subsystem page documents ctx.forge.writeReview, its write
and result types, and the panel's changeRequestReviewWrite; the package
READMEs describe the GitHub mutations and their measured behavior, the
thread actions, and the pending review flow. A new Agent Note records
moving pull request comment drafts to the code host's pending review,
and the forge panel note links it. The Web e2e replies to, edits,
deletes, and resolves threads, and starts, submits, and discards a
pending review through the scripted gh double.
Archive cleanup read whole Git listings into the 64 KiB maxOutputBytes
buffer, so every clean checkout in a repository with more than about a
thousand tracked files settled as failed.
- Drop the skip-worktree/assume-unchanged and gitlink listings; removal
follows what `git status` reports, as peer agent products do.
- Stream `git status` and stop at its first entry, and stream the
worktree listing, so neither is bounded by maxOutputBytes; restoring a
worktree shares the streamed lookup.
- Retain a checkout whose administrative directory holds submodule
repositories instead of letting Git refuse the removal.
Refs #8151
creatixchu