agentinit 1.16.0 → 1.16.2

package/CHANGELOG.md

@@ -1,3 +1,17 @@
+## [1.16.2](https://github.com/agentinit/agentinit/compare/v1.16.1...v1.16.2) (2026-04-01)
+
+
+### Bug Fixes
+
+* **skills:** harden GitHub subpath resolution ([ca353ad](https://github.com/agentinit/agentinit/commit/ca353ad46c93180b2cc0e86f89a6ea2ee53b9504))
+
+## [1.16.1](https://github.com/agentinit/agentinit/compare/v1.16.0...v1.16.1) (2026-04-01)
+
+
+### Bug Fixes
+
+* align Claude native plugin installs with marketplace behavior ([bdfa280](https://github.com/agentinit/agentinit/commit/bdfa2801a9ad933727fd72327cd9b121dfe0a543))
+
 # [1.16.0](https://github.com/agentinit/agentinit/compare/v1.15.1...v1.16.0) (2026-04-01)
 
 
@@ -142,7 +156,7 @@
 * add OpenClaw as a skills target with shared project installs via `.agents/skills/` and dedicated global installs via `~/.openclaw/skills/`
 * add Hermes as a skills target with shared project installs via `.agents/skills/` and dedicated global installs via `~/.hermes/skills/`
 * warn and fall back to GitHub repos for marketplace misses, marking allowlisted repos like `openai/codex-plugin-cc` as verified
-* install Claude Code-native plugin payloads during `plugins install` when Claude-only components are detected and Claude Code is targeted
+* install Claude Code-native plugin payloads during `plugins install` when Claude-only components are detected and Claude Code is targeted, registering native marketplace metadata under `~/.claude/plugins`
 
 ### Bug Fixes
 
@@ -152,7 +166,7 @@
 * update docs and onboarding to the `mcp add|verify`, `rules add`, and `skills add` command model
 * parse Claude Code marketplace bundle repos like `openai/codex-plugin-cc` when installing portable skills or plugins
 * keep interactive plugin target defaults on all detected agents, warn clearly when Claude-native payloads are skipped, and reuse the previewed remote source during install
-* enable Claude-native plugin payloads in `~/.claude/settings.json` so installed Claude plugins actually activate
+* enable Claude-native plugin payloads in `~/.claude/settings.json`, register them in `known_marketplaces.json`, and migrate legacy `agentinit-*` Claude namespaces to native marketplace ids
 
 # [1.7.0](https://github.com/agentinit/agentinit/compare/v1.6.0...v1.7.0) (2025-10-17)
 

package/README.md

@@ -186,6 +186,9 @@ agentinit skills add vercel-react-best-practices
 # Install all discovered skills for detected agents using canonical storage
 agentinit skills add ./skills
 
+# Install a skill stored in a repository subdirectory
+agentinit skills add owner/repo/path/to/skill
+
 # Install marketplace-hosted skills explicitly
 agentinit skills add claude/skill-creator
 agentinit skills add skill-creator --from claude
@@ -217,7 +220,7 @@ Hermes follows the same model: project skills come from `.agents/skills/`, globa
 
 Some agents share the same native skills directory. For example, Claude Code and Claude Desktop both use `~/.claude/skills/`, so `skills remove --agent ...` will skip deleting that shared path while another agent still depends on it.
 
-Bare skill names resolve from your configured default marketplace when one is set. Without a configured default, AgentInit falls back to the public skills catalog used by the open agent skills ecosystem: `vercel-labs/agent-skills`. Use `./name` for a local path, `owner/repo` for an explicit GitHub repository, or `--from <marketplace>` / `<marketplace>/<name>` for marketplace-backed sources.
+Bare skill names resolve from your configured default marketplace when one is set. Without a configured default, AgentInit falls back to the public skills catalog used by the open agent skills ecosystem: `vercel-labs/agent-skills`. Use `./name` for a local path, `owner/repo` for an explicit GitHub repository root, `owner/repo/path/to/skill` for a repository subdirectory, or `--from <marketplace>` / `<marketplace>/<name>` for marketplace-backed sources. Three-or-more slash segments are treated as a GitHub repository plus subpath, even when the owner matches a marketplace id. GitHub `tree/...` URLs and `blob/.../SKILL.md` URLs resolve to the referenced skill directory.
 
 Marketplace-backed `skills add` installs only the discovered skills. If a marketplace source also contains MCP servers or other portable components, AgentInit warns and points you to `agentinit plugins install ...` for the full install.
 
@@ -263,7 +266,7 @@ agentinit plugins remove code-review
 - Implemented marketplaces today include `claude` (Anthropic's Claude plugin marketplace) and `openai` (the OpenAI Codex skills catalog).
 - You can add your own marketplaces with `agentinit config marketplaces add <identifier> <repo-url>`.
 - If a marketplace lookup misses but the source still looks like `owner/repo`, AgentInit warns and tries that GitHub repository directly. Exact repos added with `agentinit config verified-repos add <owner/repo>` are labeled as verified during that fallback.
-- For Claude-format plugins, `plugins install` still installs portable skills and MCP servers for the selected agents. If Claude Code-native components are also present, AgentInit previews that compatibility before target selection. The native plugin payload installs only when `claude` is one of the selected targets; otherwise AgentInit warns that the Claude-only parts were skipped. When the native payload is installed, AgentInit reminds you to run `/reload-plugins`.
+- For Claude-format plugins, `plugins install` installs portable skills and MCP servers for compatible selected agents. If a native Claude Code plugin bundle is also present, AgentInit installs that bundle into `~/.claude/plugins` when `claude` is selected, registers the corresponding Claude marketplace metadata, and skips portable skill fallback for any selected agents that share Claude Code's skills directory. Portable MCP installs still apply to other compatible agents.
 - Claude-native plugin payloads are user-scoped and stored under `~/.claude/plugins`, even when the AgentInit install itself is project-scoped.
 
 ### `agentinit config`