skillset v0.16.3
skillset v0.16.3: published 2026-07-10. Package changelog notes and the original GitHub publication notice.
GitHub release · Package changelog
Publication
Published skillset@0.16.3 to npm with dist-tag latest.
Changes
Patch Changes
- 8e6a698: Add
skillset change migratefor converting legacy frontmatter pending entries to reason-only ledger entries, and demote thechange-entryschema docs to compatibility-only recovery guidance. - 8e6a698: Add schema-versioned change ledger reader support for reason-only provenance events.
- bf17d71: Extract lookup and runtime tester CLI command islands from the shared CLI core.
- 8f5fcdc: Fix
skillset testactivation probes for Cursor by treating Cursor project agents as native Markdown outputs, and addselect.agentsfor project-agent test selection. - 7e828dc: Include Cursor in the default provider target plan so omitted
compile.targetsbuilds Claude, Codex, and Cursor outputs by default. - 1463ea2: Render Cursor as a first-class provider target with native plugin, marketplace, skill, rule, agent, hook, and MCP outputs backed by provider registry evidence.
- 6543594: Complete the Cursor provider cutover across runtime tester, import/adopt, provider-format conformance, setup, lookup, Workbench diagnostics, docs, and self-hosted generated output.
- cff89f4: Add Cursor to the shared target vocabulary, schema contracts, lookup surfaces, and core provider plumbing.
- ea272e0: Add explicit
skillset dev --watch --applywrite mode that reuses build ownership, backup, and restore safeguards while keeping default watch mode read-only. - 9e2b7c6: Add inline adaptive hook runtime context delivery while reserving toolkit-backed hooks until helper provenance is ready.
- 6df1d3b: Rename the built-in provider facts package to
@skillset/registry, then add structured hook provider evidence, classifier-backed native hook lift diagnostics during import/adopt, and plugin adaptive hookrun.envrendering with deterministic shell assignments. - 4b04ab5: Add the managed XDG known-Skillsets index used to remember local checkout identities for future marketplace repository resolution without committing local filesystem paths.
- 8e6a698: Append
release.appliedledger events duringskillset release applyand derive release state from ledger evidence when present. - daab572: Support inherited Skillset license declarations, local source
LICENSE.txtfiles,license: noneopt-outs, and managed generatedLICENSE.txtoutputs for plugin bundles and skill roots. - e786cc3: Add read-only
skillset marketplace checkreadiness reports for local and known-index external marketplace entries, including generated-output freshness, target support, JSON output, and not-ready diagnostics. - 08e64b4: Add marketplace
shasource support, lock provenance, and stale/pinned readiness diagnostics forskillset marketplace check. - 17c221d: Add
skillset marketplace updateto render provider-supported marketplace indexes from Skillset marketplace source, refresh marketplace provenance in the existing rootskillset.lock, and support mixed local/external Claude catalog entries through derived git-subdirectory sources. - 6df1d3b: Keep native hook companion JSON files such as
hooks/hooks-cursor.jsonfrom being misclassified as adaptive hook units during plugin adoption. - 7ba60b7: Retire app-level core re-export facades and add a package ownership guard so CLI modules import owned package APIs or documented package internals directly.
- 02dfdfa: Move default generated plugin output to the plugin-first
plugins/<plugin>/<target>/layout, with root-level Claude marketplace output and compatibility for explicit provider output roots. - d542328: Replace the retired
tool_intentcontract with portabletoolspolicy frontmatter, including provider-native overrides, schema validation, lookup docs, and generated target metadata. - 85e6a5c: Document and prove project-agent orchestration compatibility by tying Claude native agent skill metadata and Codex skill-loading prefaces to activation probe evidence.
- 8e6a698: Cut change authoring commands over to reason-only pending entries backed by ledger events.
- f1f8a4a: Stop scaffolding tracked
.skillset/cache/.gitignoresentinels now that operational cache payloads are XDG-backed. - 771eba8: Stop treating retired provider-first plugin output roots as importable or conformance-relevant paths now that generated plugin output uses the plugin-first
plugins/<plugin>/<provider>layout. - 886998f: Share retained-run cache plumbing between
skillset testandskillset runtime-tester. - ed41dce: Fail loudly when
skillset runtime-tester --pluginnames an unknown or unsupported plugin. - 816f3cb: Add
skillset runtime-testerfor retained non-interactive Claude and Codex prompt runs against isolated generated output, with Claude setting sources configurable by CLI flag or environment variable. - da413f7: Render adaptive hooks with
context.strategy: toolkitthrough an internalskillset hooks contextruntime helper, proving helper-backed context delivery before the public toolkit package boundary lands. - 1caebf3: Render adaptive hook
context.strategy: toolkitwrappers through the publishedskillset-toolkitruntime context bin. - 8b7f0f5: Add a shell-friendly
@skillset/toolkitruntime context CLI for JSON and env output. - 49dad00: Document and verify the toolkit runtime context support matrix through lookup output, schema examples, and adaptive hook fixtures.
- e2cbe8b: Promote the runtime context helper into the private
@skillset/toolkit/runtimeworkspace package so hook scripts can build against the same typed provider context model the CLI uses. - e0918ae: Add the portable tools capability registry and realization planner: registry facts record realization tier, surface, emitted field/rule family, diagnostics, and evidence per provider and aspect; renderers, render results,
skillset lookup skill tools --compat, andskillset explainconsume those facts; Cursor skills now receive the.skillset.tools.yamlmetadata sidecar and Cursor tools overrides are validated at lint/build. - 571d094: Enable
compile.unsupportedDestination: warn,skip, andforcefor unsupported or lossy render results while failed render results still block every policy and selected soft policies remain visible in diagnostics plus lock provenance.