Skip to content

Releases: PerryLink/dsh-plugin-kit

v0.1.20

Choose a tag to compare

@github-actions github-actions released this 18 Oct 14:18

[0.1.20] - 2026-10-18

Changed

  • Move tsdown/typescript (and @types/react where present) from runtime dependencies to devDependencies. Git installs still build: pnpm installs devDependencies for a git dependency and runs prepare; npm tarballs ship the built lib/. The runtime tree no longer carries the build chain.

v0.1.19

Choose a tag to compare

@github-actions github-actions released this 18 Oct 09:38

[0.1.19] - 2026-10-18

Changed

  • Add a 1280×720 hero and a 960×540, 4-second terminal GIF rendered from the test run; screenshots.json is an object array and capture.json records the command.

v0.1.17

Choose a tag to compare

@github-actions github-actions released this 10 Oct 01:08

[0.1.17] - 2026-10-10

Changed

  • The README now carries the terminal demo and an animated version of it: the same
    real run, shown assembling line by line. The static image stays as the primary
    one, because a crawler or an assistant can read a PNG and cannot read a GIF.

    Both frames are that plugin's own output — the card is generated by running its
    code over its own fixture, not drawn by hand.

v0.1.16

Choose a tag to compare

@github-actions github-actions released this 05 Oct 14:02
032174a

[0.1.16] - 2026-10-05

Fixed

  • verify-seam no longer fails a plugin that carries no capability seam. The gate reported
    a false positive whenever none of the three role markers was present, which is the normal
    shape of a pure detector or a read-only tool: it registers nothing for anyone else to
    consume, so there is no incomplete trio to report. dsh-plugin-doctor — a flat .mjs +
    lib/ repo with no src/ — failed node lib/verify/cli.js all . for exactly this reason,
    and so did the doctor own self-check, which cannot be red on its own criteria.
    The gate now measures the condition instead of assuming it: it fires only once at least one
    role marker is present, and reports the out-of-scope case as a warning so the outcome stays
    visible rather than indistinguishable from a fully conforming trio. The failure mode the gate
    exists to catch — a half-declared seam — still fails, with both the two-of-three case and the
    root-level pure-JS fallback pinned by tests.

Notes

  • No behaviour change for any repository that carries a seam: the three-role assertion is
    byte-identical to before. Only the no-seam case changed, from FAIL to PASS + WARN.

v0.1.15

Choose a tag to compare

@github-actions github-actions released this 05 Oct 11:46
e8104e8

[0.1.15] - 2026-10-05

Changed

  • Correct the release date in the previous section, which was stamped with the literal string undefined by the release stamper. No content or behaviour change; the version is bumped only because npm will not republish an existing version.

v0.1.13

Choose a tag to compare

@PerryLink PerryLink released this 04 Oct 18:03
a332adf

[0.1.13] - 2026-10-04

Changed

  • Host pins move to 0.2.1-alpha.1; re-verified against that host line. Every @deepseek-ai/dsh-* dev/test dependency now pins 0.2.1-alpha.1, the dshWorkshop.compatibility.dshVersions timeline appends 0.2.1-alpha.1, and the compatibility baseline in every README records the dsh-v0.2.1-alpha.1 host. The declared host ranges (engines.dsh and the peerDependencies union) gain the || >=0.2.0-0 <0.3.0 || >=0.2.1-0 <0.3.0 clauses: the previous upper bound was <0.2.0, which under semver rejects every 0.2.x host, so the probe host itself was not installable. Nothing was narrowed — the 0.1.x clauses are unchanged, in place and in order.

v0.1.9

Choose a tag to compare

@github-actions github-actions released this 12 Sep 02:31

[0.1.9] - 2026-09-12

Added

  • scripts/check-peer-range-latest.mjs — a dependency-free tripwire that reads the published @deepseek-ai/dsh dist-tags and fails when the canonical peer range no longer admits one, printing the clause to append. semver only admits a prerelease when a comparator in the same [major, minor, patch] tuple carries a prerelease, so every new upstream prerelease tuple silently stops matching; this turns that into a red gate. Wired into .github/workflows/peer-range.yml (push, PR, nightly, manual).

Changed

  • Rename the four translated READMEs to README-<lang>.md. npm selects the package-page readme as the first markdown file matching its {README,README.*} glob (@npmcli/package-json, publish path), and that glob order puts README.<lang>.md ahead of README.md — so npm was serving the Simplified-Chinese file for this package too (measured on 15/15 sampled packages of the family). The new names sit outside the glob, so the English source is served again. No content changed apart from the language-switcher link each translation holds to its siblings, and the repo readme gate still passes. Takes effect with the next release; an already-published version cannot gain a corrected readme retroactively.
  • data/peer-range.json canonical range is now >=0.1.2-rc.1 <0.2.0 || >=0.1.5-alpha.1 <0.2.0 (was the stale >=0.1.0-rc.8 <0.2.0 from 2026-08-26), which is what the ecosystem already declares.
  • scripts/sync-peer-range.mjs understands the ||-joined clause form: parseRangeSet/formatRangeSet were added and rangeStatus/targetRange now compare clause sets instead of a single floor/upper pair. The single-clause parseRange keeps its old signature, so existing callers and tests are unaffected.
  • The release workflow now creates the GitHub Release itself, with the body taken from this version's CHANGELOG section. Until now a v* tag published to npm and stopped there, so every Release page had to be created by hand afterwards.
  • Pin the @deepseek-ai/dsh-* dev/test dependencies to the published 0.1.5-rc.2 line; the monthly Compat workflow now runs against 0.1.5-rc.2. The peer range >=0.1.2-rc.1 <0.2.0 || >=0.1.5-alpha.1 <0.2.0 is unchanged, so no supported host line is dropped.

Fixed

  • verifyReadmeLanguages now expects README-<lang>.md. It built the dotted name from a template, so a textual rename cannot see it — this is the one place in the fleet that constructs the filename dynamically, and its own test is what caught the mismatch. Nothing consumes this verifier yet (each repo carries its own verify-readmes copy), so the change stays inside this repo.

v0.1.8

Choose a tag to compare

@PerryLink PerryLink released this 10 Sep 06:46

[0.1.8] - 2026-09-10

Changed

  • Pin the @deepseek-ai/dsh-* dev/test dependencies to the published 0.1.5-rc.1 line and record 0.1.5-rc.1 in dshWorkshop.compatibility.dshVersions; the monthly Compat workflow now runs against 0.1.5-rc.1. The peer range >=0.1.2-rc.1 <0.2.0 || >=0.1.5-alpha.1 <0.2.0 is unchanged, so no supported host line is dropped.

Docs

  • Refresh the five-language README compatibility baseline to dsh-v0.1.5-rc.1 (verified 2026-09-10).

v0.1.7

Choose a tag to compare

@PerryLink PerryLink released this 09 Sep 14:58

[0.1.7] - 2026-09-09

Changed

  • Align the @deepseek-ai/dsh-* peer ranges to >=0.1.2-rc.1 <0.2.0 || >=0.1.5-alpha.1 <0.2.0 and pin the dev/test dependencies to the published 0.1.5-alpha.1 line: adaptation to DeepSeek Harness dsh-v0.1.5-alpha.1 (session format V3, ctx.agent removal, Inbox type-only interface); runtime behavior is unchanged for every supported host line.
  • Record 0.1.5-alpha.1 in dshWorkshop.compatibility.dshVersions.

Docs

  • Refresh the five-language README compatibility baseline to dsh-v0.1.5-alpha.1 (verified 2026-09-09).

v0.1.6

Choose a tag to compare

@PerryLink PerryLink released this 07 Sep 10:54

[0.1.6] - 2026-09-07

Fixed

  • Align the @deepseek-ai/dsh-* peer ranges to >=0.1.2-rc.1 <0.2.0: the older >=0.1.0-rc.8 <0.2.0 band resolved to only the 0.1.0-rc.8 prerelease under registry-driven resolution and broke fresh tarball installs; no behavior change.

Docs

  • Refresh the five-language README support-version wording: the verified GitHub tag dsh-v0.1.3-alpha.1 now leads the compatibility claim, while npm 0.1.2-rc.1 stays the published dependency-pin line (peers >=0.1.2-rc.1 <0.2.0); no behavior change.