ChatGPTClaudeGrok / X / CursorGemini

DeepSeek Harness alpha Install ETARGET: Missing Leaf Package

TroubleshootingPublished 2026-11-03Author: DeepSeek Plugin Market
DeepSeek HarnessETARGETalpha channelnpm install failuredist-tagleaf package
npm i -g @deepseek-ai/dsh@alpha fails with ETARGET because 0.2.2-alpha.3 pins a leaf package never published. Diagnosis, release view, recovery timeline.

The DeepSeek Harness (DSH) alpha channel was once uninstallable end to end: both npm i -g @deepseek-ai/dsh@alpha and npx @deepseek-ai/dsh@alpha failed with ETARGET during dependency resolution, reporting No matching version found for @deepseek-ai/dsh-tool-pwsh-persistent@0.2.2-alpha.3. This is not a cache or npmrc problem — npm i --dry-run fails identically because the install never reaches the write step. What was missing is a leaf package that was never published, and the main package pinned it to an exact version, so a single gap made the channel unusable for every user (#11984).

Three commands to identify which package is missing

The decisive signals: ETARGET appears during resolution, the error names a specific package and version, and switching to latest works immediately. Run these in order.

  1. See where the channel tag currently points.
bash
npm view @deepseek-ai/dsh dist-tags
#   { alpha: '0.2.2-alpha.3', latest: '0.2.1-rc.5', next: '0.2.1-rc.5' }

Expected: alpha points at a concrete version. This step fixes "which version am I even installing" and keeps "the tag points at the wrong thing" separate from "that version has a missing dependency".

  1. Confine the failure to resolution with --dry-run.
bash
npm i --dry-run @deepseek-ai/dsh@0.2.2-alpha.3
#   npm error notarget No matching version found for
#   @deepseek-ai/dsh-tool-pwsh-persistent@0.2.2-alpha.3

Expected: the same ETARGET and no install actions at all. The value of --dry-run is proving the failure happens in resolution rather than download or disk writes, which rules out network interruptions and disk permissions.

  1. Query the named leaf package directly.
bash
npm view @deepseek-ai/dsh-tool-pwsh-persistent versions --json
npm view @deepseek-ai/dsh-tool-pwsh-persistent dist-tags

Expected: 0.2.2-alpha.3 is absent from versions, and its alpha tag still sits at 0.2.2-alpha.2 (published 2026-10-27T19:03:34Z). At this point you know exactly what is missing and for how long.

One supporting data point: 0.2.2-alpha.3 declares 90 dependencies with 1 unresolvable; 0.2.2-alpha.2 declares 91 with 0 unresolvable; 0.2.1-rc.5 declares 86 with 0 unresolvable. This was not a half-finished release; it was exactly one package short (#11984).

Why one leaf package takes down the whole channel

Short answer: exact pins amplify "one package missing" into "the whole tree cannot resolve", and npm resolution is all-or-nothing. The mechanism has three layers.

  1. The main package declares exact versions. @deepseek-ai/dsh@0.2.2-alpha.3 writes sibling packages as full version strings like 0.2.2-alpha.3 rather than ^ or ~ ranges.
  2. npm requires a matching version for every exact dependency. If even one cannot be found, resolution fails wholesale; npm does not "skip that package and keep installing".
  3. The failure surfaces at the outermost layer. A missing version anywhere in the tree reports back through the npm i you ran, so it looks like "DSH will not install" when in fact an internal package was never published.

That also explains why users had no self-service fix: ETARGET is a resolution-time error, there is no "patch it after half the install" path, and the only options are waiting for a republication or switching versions.

The full timeline of this incident

Short answer: the gap was closed by an after-the-fact republication, roughly 29 minutes from detection to recovery. In UTC:

  1. 07:48:26Z: @deepseek-ai/dsh@0.2.2-alpha.3 lands in the registry, and the alpha tag moves to it.
  2. Around 08:10Z: the reporter checks the registry and confirms dsh-tool-pwsh-persistent still has 0.2.2-alpha.2 as its newest version, with no alpha.3 in versions.
  3. 08:44Z: another community member queries the registry directly package by package (bypassing local caches) and independently confirms the same conclusion: the alpha channel is uninstallable for everyone, the only gap is that leaf package, the other 88 exact pins all exist, and all 8 range declarations have a satisfying version.
  4. 09:03:15Z: the republication lands. dsh-tool-pwsh-persistent now reports alpha: '0.2.2-alpha.3' and the channel recovers.
  5. Real-machine verification after the republication (not a dry run):
bash
npm install @deepseek-ai/dsh@alpha
#   added 652 packages in 3m
#   EXIT=0

Expected: the install succeeds, both node_modules/@deepseek-ai/dsh and node_modules/@deepseek-ai/dsh-tool-pwsh-persistent point at 0.2.2-alpha.3, and the 90 dependencies now show zero missing.

One detail worth remembering from the timeline: the release script is strictly serial, dependency-topology ordered, and fail-fast (publishRelease in scripts/release/publish.ts), so the main package is necessarily published after its leaves — and the main package did land. In other words the pipeline did not stop mid-way with an error; it looks more like this package never entered the packaging/upload set for this wave. That can only be inferred from outside the repo, but it is enough to show that "the pipeline succeeded" does not mean "every member reached the registry".

One common misread: 91 dependencies dropping to 90 is not evidence of a partial release

Short answer: dsh-cua-bundle disappearing from the declaration is a real difference in the tag tree, not a missing-publish signal; using it as evidence leads to the wrong conclusion. The thread initially listed it as supporting evidence for "only half the release wave shipped", and an independent review corrected that.

  1. Confirm the source of the difference with a tag-tree diff. Diffing apps/cli/package.json between the two tags shows the line "@deepseek-ai/dsh-cua-bundle": "workspace:*" was removed.
  2. Understand what that package is. cua-bundle is an optional bundle with dsh.bundle.platforms: ['darwin'], and the commit that introduced it is not an ancestor of alpha.3, so the tag never declared it.
  3. Confirm the key package is declared in both tags. pwsh-persistent exists in both the alpha.2 and alpha.3 trees, so the main conclusion of "exactly one gap" stands.
  4. Correct the conclusion: detecting a missing package requires comparing each dependency against the registry, never the total dependency count and never the presence of some optional bundle.

The same thread contains another output that is easy to mistake for signal: npm error config prefix cannot be changed from project config: ~/.npmrc is a local npmrc configuration issue and has nothing to do with ETARGET.

What the release side should gate on: do not wait for users to hit ETARGET

Short answer: the real gap is the release pipeline — its verification installs local tarballs, which cannot detect "present locally, absent from the registry". The discussion suggests adding a post-publish assertion.

  1. After npm publish, query every member of the family (npm view <name>@<version> version). Expected: every member responds; if any does not, abort and do not move the dist-tag.
  2. Or make "every exact dependency resolves from the registry" a pre-publish gate. Run npm view on each entry in the target version's dependencies and only tag the release when all of them hit.
  3. Consider switching sibling packages to ~ / ^ ranges. A partially published wave would then resolve to the nearest published version instead of failing every install. The thread notes this was already the third consecutive time an exact pin on an unpublished sibling broke the channel.
  4. Keep release:verify-packed-install, but do not rely on it alone. It installs a tree from local tarballs, which proves local packages can install each other; it proves nothing about registry visibility.

This is not a one-off: dsh-client-ui-sidebar-documentpreview@0.1.5-rc.3 in 0.1.5-rc.*, dsh-client-ui-settings-account in 0.2.0-rc.2, and dsh-compaction-basic in 0.1.0-rc.6 all belong to the same family — the main package shipped first and the leaf did not follow (#11984).

What to do on the user side now

Short answer: first check whether the republication is complete; if not, move to latest or a previous alpha, and if it is, install alpha directly. Steps:

  1. Compare the two tags to see whether the gap still exists: npm view @deepseek-ai/dsh dist-tags and npm view @deepseek-ai/dsh-tool-pwsh-persistent dist-tags. Expected: once the latter's alpha points at 0.2.2-alpha.3, the gap is closed and you can install.
  2. When the republication is complete, install alpha directly: npm i -g @deepseek-ai/dsh@alpha.
  3. When the gap is still open, install the previous alpha (all 91 dependencies resolve): npm i -g @deepseek-ai/dsh@0.2.2-alpha.2.
  4. Or take the stable line: npm i -g @deepseek-ai/dsh@latest (0.2.1-rc.5, all 86 dependencies resolve).
  5. Do not try to "fix" ETARGET by clearing caches, switching registry mirrors, or reinstalling Node. None of those changes the resolution outcome.
  6. Verify the actual versions after installing: npm ls -g @deepseek-ai/dsh @deepseek-ai/dsh-tool-pwsh-persistent. Expected: both packages report the same 0.2.2-alpha.3.

Do not overlook the follow-on problem "it installed but will not run": once 0.2.2-alpha.3 installs, the alpha line still has other issues such as Agent presets failing to load, covered in DeepSeek Harness Agent presets all fail to load.

Things to keep in mind while debugging

The principle for ETARGET-style failures: separate "failed during resolution" from "failed after writing to disk", then read the package named in the error. Six points.

  1. ETARGET / notarget means the registry has no such version — not a network issue, not a permission issue, not a cache issue.
  2. --dry-run is the fastest triage tool: if it fails too, the problem has nothing to do with your local environment.
  3. Read the package name and version in the error, not just the outer "DSH will not install" description.
  4. Compare each dependency against the registry; never treat a changed dependency count or the presence of an optional bundle as evidence.
  5. A dist-tag move does not mean dependencies are complete; a tag is just an alias with no validation semantics.
  6. State fix status precisely: this case was fixed by a republication at 2026-11-03T09:03:15Z and verified with a real install.

Source: Discussion #11984.

ChatGPTClaudeGrok / X / CursorGemini

FAQ

DeepSeek Harness alpha install ETARGET: is my npm cache bad?

No — this DeepSeek Harness ETARGET is thrown during dependency resolution, which means the registry genuinely lacks that version; local cache and npmrc settings are irrelevant. npm i --dry-run reproduces the same failure without writing anything, because it never gets past resolution.

One missing leaf breaks DeepSeek Harness alpha installs?

Because the DeepSeek Harness main package declares its sibling packages with exact version pins, and npm requires every exact dependency to resolve. If one is missing, the whole resolution aborts. 0.2.2-alpha.3 declares 90 dependencies; 89 resolve and only dsh-tool-pwsh-persistent@0.2.2-alpha.3 does not exist, which takes down the whole alpha dist-tag.

Can DeepSeek Harness 0.2.2-alpha.3 be installed now?

Yes — the DeepSeek Harness ETARGET issue is fixed: a community member confirmed in the thread that dsh-tool-pwsh-persistent@0.2.2-alpha.3 was republished at 2026-11-03T09:03:15Z, the alpha dist-tag points back at 0.2.2-alpha.3, and all 90 dependencies resolve. A real install of npm install @deepseek-ai/dsh@alpha succeeded with added 652 packages in about 3 minutes.

Did DeepSeek Harness publish only half of 0.2.2-alpha.3?

No — in DeepSeek Harness that difference comes from the git tag tree itself: dsh-cua-bundle is an optional bundle declared only for darwin, and the dsh-v0.2.2-alpha.3 tag never declared it, so its absence says nothing about release completeness. To detect a missing package, compare each dependency against the registry; never use the total dependency count as a signal.

Related Terms

ETARGET
ETARGET is the error code npm raises during dependency resolution when a requested package version has no match in the registry, usually phrased No matching version found for <package>@<version>. It happens before anything is written to disk, so it never leaves a half-finished install behind.— https://github.com/deepseek-ai/deepseek-harness/discussions/11984
dist-tag
A dist-tag is an alias npm attaches to a package version, such as latest, next, or alpha; installing with @alpha resolves through that tag. The tag itself does not verify that every dependency of its target version resolves, so a tag can point at a version nobody can install.— https://github.com/deepseek-ai/deepseek-harness/discussions/11984
exact pin
An exact pin writes the full version string in a dependency declaration (for example 0.2.2-alpha.3) instead of a range such as ^ or ~. The cost is that any sibling package missing from the publish wave fails the entire tree, while a range declaration could fall back to the nearest published version.— https://github.com/deepseek-ai/deepseek-harness/discussions/11984

Sources