Skip to content
Blume
Esc
↑↓navigate↵open⌘Jpreview
Open Markdown
On this page

skillset v0.27.0

skillset v0.27.0: published 2026-09-18. Package changelog notes and the original GitHub publication notice.

GitHub release · Package changelog

Publication

Published the seven-package Skillset 0.27.0 set to npm with dist-tag latest.

Changes

Minor Changes

  • 545a73f: Adopt the validated Agent Instructions, Agent Skills, and Agent Plugins profiles so default builds emit their inherent standards projections with pinned conformance evidence.
  • b0f6b1c: Remove standards selection from source configuration and add Agent Skills compatibility plus allowed_tools.agents. Standards remain distinct from provider targets, and an explicit empty compile.targets selection is available for applicable adopted standards output. Normal builds gain each inherent applicable standard when its profile is adopted; pre-v3 generated locks require an owner-reviewed one-time rebuild and never authorize cleanup.
  • 21940a2: Generate individually installable Agent Skills from self-contained plugin source while diagnosing plugin dependencies and skill-local hooks that cannot survive standalone installation.
  • ec70bf5: Generate one repository-owned ChatGPT marketplace catalog from ordered Skillset marketplace source, with closed native source, policy, listing fallback, and interface mappings.
  • b863f40: Render the Codex-selected ChatGPT product bundle as an Agent Plugins root with a closed OpenAI extension, portable Agent Skills and MCP components, deterministic ownership locks, and pinned OpenAI consumer evidence.
  • 774b9de: Add a public registry surface for immutable candidate Agent Instructions, Agent Skills, and Agent Plugins 1.0 evidence, including offline plugin and MCP schema snapshots and report-only drift checks.

Patch Changes

  • e792ac3: Render Agent Instructions independently from provider selection, coalesce shared Codex AGENTS.md output under one standard owner, and honor explicit Agent Instructions opt-outs without recreating provider fallback files.
  • ccc667b: Render Agent Plugins 1.0 package manifests and neutral support files under plugins/<plugin>/agents, preserve standard ownership in the shared lock, report provider-only components as uncovered without leaking them into the standard package, and reject symlinked or out-of-scope license inputs before reading them.
  • d25e439: Render Agent Skills baselines for standalone and plugin-owned skills, coalesce compatible Codex skill trees with standard ownership, reject resource symlink escapes, and report standard-invalid components without suppressing valid provider output.
  • 16830c0: Fix implicit Claude marketplace catalogs so local plugin sources are not rebased twice by consumers.
  • cb9704c: Render canonical listing display names into Claude plugin manifests while preserving provider-native and marketplace label precedence.
  • 7c5143a: Coalesce standard baselines and provider deltas into one deterministic physical output plan, and safely clean inactive standard roots across live and isolated builds.
  • efe5cc7: Render versioned, flat Cursor plugin hook handlers from existing grouped hook source, preserve validated Cursor-native handler fields, omit portable status text that has no Cursor destination, and reject malformed flat handlers that previously bypassed validation.
  • 24bb38b: Lower canonical implicit_invocation to Cursor disable-model-invocation with inverse polarity, preserve native override precedence, and scope adopted provider-native invocation policy so it cannot leak into unrelated targets.
  • 953809b: Read current generated locks and render results through validated, versioned readers while preserving lock identity and logical-consumer evidence. Diagnose pre-v3 generated state as a coordinated rebuild instead of granting it cleanup authority.
  • 1c2c8cb: Import Agent Plugins 1.0 packages into canonical plugin source, including portable metadata, skills, MCP configuration, and recognized support files, while blocking unmappable extensions before adoption writes.
  • ebadb1f: Document Agent standards as inherent adopted projections, distinguish project-agent and provider configuration from standards, enforce Agent Skills names in skillset new skill, and clarify the generated CI scaffold.
  • 6008bf9: Add version-qualified provider location evidence for distinct Claude, Codex, Cursor, and ChatGPT surfaces and expose it through the static lookup API and CLI.
  • c338af7: Refresh the exact hosted provider validator evidence and expose receipt-based freshness assessment for maintenance tooling.
  • 1926f66: Keep command-only build mode overrides out of generated lock provenance so later default status, check, diff, and update commands remain current.
  • 0c783cf: Expose standard-profile lifecycle, scope, evidence, ownership, and consumer identity across CLI and JSON inspection surfaces while keeping provider activation readiness separate from standards conformance.
  • 72447c0: Derive inherent Agent standards projections from applicable source and adopted registry profiles, independently from provider target selection.
  • f5fb227: Track SEP-2640 as an immutable candidate standard and record OpenAI’s separate plugin-submission skill importer limits.
  • 20bd533: Parse plugin MCP source into one portable model, validate the Agent Plugins contract, render provider bundle bytes from that model, emit standard mcp.json, and rewrite provider dialects during import.

Was this page helpful?