agentera 3.0.0-dev.2 → 3.0.0-dev.21

package/bundle/references/adapters/opencode.md

@@ -1,6 +1,6 @@
 # OpenCode Adapter
 
-Maps agentera's host adapter contract (SPEC.md Section 21) and session corpus contract (SPEC.md Section 22) to OpenCode's specific mechanisms. A developer reading only this document can implement portable-core agentera support and a profilera-compatible session corpus in OpenCode without reading any skill entry file source code.
+Maps agentera's host adapter contract (SPEC.md Section 21) and session corpus contract (SPEC.md Section 22) to OpenCode's specific mechanisms. A developer reading only this document can implement portable-core agentera support and a profile-compatible session corpus in OpenCode without reading any skill entry file source code.
 
 ---
 
@@ -17,7 +17,7 @@ Status: production reference (formerly design document).
 | Skill install | Documented below | `npx skills add jgabor/agentera -g -a opencode --skill agentera -y` |
 | Profile path | Documented below | `~/.config/opencode/profile/PROFILE.md` |
 
-Routes exact bare text hej through chat.message to the bundled Agentera dashboard path. The public docs page omits this hook, but `packages/plugin/src/index.ts` defines `"chat.message"` on the exported `Hooks` interface.
+Routes exact bare text /agentera through chat.message to the bundled Agentera dashboard path. The public docs page omits this hook, but `packages/plugin/src/index.ts` defines `"chat.message"` on the exported `Hooks` interface.
 
 The plugin was promoted from `references/adapters/opencode-plugin.js` to `.opencode/plugins/agentera.js` and is now the production local-file location. Install it with:
 
@@ -56,13 +56,14 @@ OpenCode skill search paths:
 - Global install: `npx skills add jgabor/agentera -g -a opencode --skill agentera -y`
 - Project install: symlink or copy `skills/agentera/` into `.opencode/skills/agentera/` or `.agents/skills/agentera/`
 
-The temporary `skills/hej/` entry point is a v1 upgrade bridge, not a new-install target. Each Agentera skill entry file already contains YAML frontmatter with `name` and `description`, which matches OpenCode's frontmatter requirements exactly. The OpenCode plugin adds one prompt transformation: when the complete user message is exactly bare text `hej`, `chat.message` rewrites it to load the bundled `agentera` skill and route through the `agentera hej` dashboard path. The match is exact except for OpenCode's CLI-added single trailing newline transport artifact, and does not apply to `/hej`, `Hej`, `hej there`, attachments, or other message parts.
+Each Agentera skill entry file contains YAML frontmatter with `name` and `description`, which matches OpenCode's frontmatter requirements exactly. The OpenCode plugin adds one prompt transformation: when the complete user message is exactly bare text `/agentera`, `chat.message` rewrites it to load the bundled `agentera` skill and route through the `agentera prime` dashboard path. The match is exact except for OpenCode's CLI-added single trailing newline transport artifact, and does not apply to `/agentera plan`, `/agentera status`, attachments, or other message parts.
 
 **OpenCode frontmatter validation** requires: `name` (1-64 chars, lowercase alphanumeric with single hyphens), `description` (1-1024 chars). Agentera entry point names are lowercase, matching the validation regex `^[a-z0-9]+(-[a-z0-9]+)*$`.
 
 **Skill loading**: OpenCode loads skills on-demand via a native `skill` tool. Agents see available skills listed in the tool description and load full content by calling `skill({ name: "agentera" })`. Agentera works unmodified: it is loaded into the agent's context when invoked.
 
-**Gap**: None. OpenCode's skill discovery is more flexible than Claude Code's (supports `.opencode/`, `.claude/`, and `.agents/` paths). Agentera skills install cleanly.
+**Gap**: None. OpenCode skill discovery supports `.opencode/`, `.claude/`
+compatibility, and `.agents/` paths. Agentera skills install cleanly.
 
 ### Artifact resolution (Required)
 
@@ -82,14 +83,14 @@ The temporary `skills/hej/` entry point is a v1 upgrade bridge, not a new-instal
 
 **Adapter approach**: Place PROFILE.md at `~/.config/opencode/profile/PROFILE.md` (or `~/.config/opencode/PROFILE.md` for simplicity). Update the profile-path references:
 
-- Skills reference `$PROFILERA_PROFILE_DIR/PROFILE.md` (default: `$XDG_DATA_HOME/agentera/PROFILE.md`) with `<!-- platform: profile-path -->` annotations
+- Skills reference `$AGENTERA_PROFILE_DIR/PROFILE.md` (default: `$XDG_DATA_HOME/agentera/PROFILE.md`) with `<!-- platform: profile-path -->` annotations
 - The OpenCode adapter substitutes `~/.config/opencode/profile/PROFILE.md`
-- Profilera writes to this path when generating the profile
+- Profile writes to this path when generating the profile
 
 **Concrete substitution**: In contract.md files, the annotated line:
 
 ```
-Read PROFILE.md from the profilera-determined profile path (`$PROFILERA_PROFILE_DIR/PROFILE.md`, defaulting to `$XDG_DATA_HOME/agentera/PROFILE.md` on Linux). <!-- platform: profile-path -->
+Read PROFILE.md from the profile-determined profile path (`$AGENTERA_PROFILE_DIR/PROFILE.md`, defaulting to `$XDG_DATA_HOME/agentera/PROFILE.md` on Linux). <!-- platform: profile-path -->
 ```
 
 becomes in the OpenCode context:
@@ -112,14 +113,14 @@ Read PROFILE.md from the runtime-provided profile path (Section 21). In OpenCode
 
 **Adapter approach**: Map agentera's worktree isolation to one of two strategies:
 
-**Strategy A: Managed descriptor dispatch (recommended for current port)**
+**Strategy A: Single agent dispatch (D73)**
 
-- Define Agentera's dispatched capabilities as OpenCode subagents
-- Each capability gets a managed markdown file in `.opencode/agents/`
-- Descriptor frontmatter stays within documented OpenCode agent fields: `description` and `mode: subagent`; the file name supplies the agent name.
-- Agentera ownership is recorded in a body comment marker, not custom frontmatter, so unknown options are not passed through as model options.
-- The OpenCode plugin copies those descriptors to `~/.config/opencode/agents/` on startup and preserves user-owned collisions
-- The orchestrating skill (realisera, orkestrera) uses `@<capability>` to invoke work
+- A single Agentera primary agent (`agentera.md`) in `.opencode/agents/`
+- The agent loads the Agentera skill for routing, runs `agentera prime --context <capability> --format json` to fetch instructions, then dispatches to a built-in `general` subagent
+- Descriptor frontmatter uses `description` and `mode: primary` with broad `permission` (write+bash allow)
+- Agentera ownership is recorded in a body comment marker
+- The OpenCode plugin copies the agent descriptor to `~/.config/opencode/agents/` on startup
+- The orchestrating skill (build, orchestrate) uses `@agentera` to invoke work
 - Limitation: runs in the same working tree, not isolated. Suitable for non-destructive work.
 
 **Strategy B: Manual git worktree (full isolation parity)**
@@ -130,18 +131,18 @@ Read PROFILE.md from the runtime-provided profile path (Section 21). In OpenCode
 - Provides true parallel implementation with independent branches
 - Requires explicit orchestration in the skill workflow
 
-**Concrete mapping for realisera Step 5 dispatch**:
+**Concrete mapping for build Step 5 dispatch**:
 
 ```
-# Claude Code (reference)
-Spawn a Sonnet implementation agent in a worktree (isolation: "worktree")
+# Runtime-neutral requirement
+Spawn an implementation agent in an isolated worktree
 
 # OpenCode Strategy A
-Invoke the relevant managed capability descriptor, for example @realisera, with the implementation plan
+Invoke the Agentera agent (@agentera) with the implementation plan
 
 # OpenCode Strategy B
 git worktree add ../worktree-branch branch-name
-Dispatch the relevant managed descriptor with cwd set to ../worktree-branch
+Dispatch @agentera with cwd set to ../worktree-branch
 After completion: git merge, git worktree remove
 ```
 
@@ -167,7 +168,8 @@ opencode run "Explain closures in JavaScript"
 opencode run --format json "Explain closures in JavaScript"
 ```
 
-The `--format json` flag produces a raw JSON event stream, comparable to `claude -p --output-format json`. Each event is a separate JSON object representing a message part, tool call, or status update.
+The `--format json` flag produces a raw JSON event stream. Each event is a
+separate JSON object representing a message part, tool call, or status update.
 
 For long-running sessions, OpenCode's server mode provides an HTTP API with the same structured output:
 
@@ -176,15 +178,8 @@ opencode serve --port 4096
 # Then POST to the API with session creation and message sending
 ```
 
-**Adapter approach**: Map `claude -p --output-format json` to `opencode run --format json`:
-
-| Claude Code | OpenCode |
-|-------------|----------|
-| `claude -p "prompt"` | `opencode run "prompt"` |
-| `claude -p --output-format json` | `opencode run --format json "prompt"` |
-| `claude -p --skill realisera` | `opencode run "prompt"` (skill loaded via discovery) |
-
-The agentera eval runner (`scripts/eval_skills.py`) would need an OpenCode dispatch mode that calls `opencode run --format json` instead of `claude -p --output-format json`. The dispatch wrapper:
+**Adapter approach**: The eval runner invokes `opencode run --format json`
+directly and parses OpenCode's event schema. The dispatch wrapper:
 
 ```python
 def dispatch_opencode(skill_name: str, prompt: str) -> dict:
@@ -195,7 +190,9 @@ def dispatch_opencode(skill_name: str, prompt: str) -> dict:
     return {"output": result.stdout, "exit_code": result.returncode}
 ```
 
-**Gap**: Minimal. `opencode run --format json` provides structured JSON events directly comparable to `claude -p --output-format json`. The eval runner adapter parses the JSON event stream using the same pattern as the Claude Code adapter, though the event schema differs and requires a separate parser module. For smoke tests (the eval runner's primary use case), text output with exit-code checking is also sufficient.
+**Gap**: Minimal. `opencode run --format json` provides structured JSON events.
+The eval runner uses a dedicated OpenCode parser. For smoke tests, text output
+with exit-code checking is also sufficient.
 
 ### Hook lifecycle (Optional but recommended)
 
@@ -207,10 +204,10 @@ def dispatch_opencode(skill_name: str, prompt: str) -> dict:
 |---------------|----------------|-------|
 | SessionStart | `event` hook with `event.type === "session.created"` | Observation only unless a supported context-injection hook is also used |
 | Stop | `event` hook with `event.type === "session.idle"` | Fires when a session reaches idle state |
-| Exact bare `hej` | `chat.message` | Rewrites only the complete text message `hej` so the bundled `agentera` skill owns the dashboard path |
+| Exact bare `/agentera` | `chat.message` | Rewrites only the complete text message `/agentera` so the bundled `agentera` skill owns the dashboard path |
 | PreToolUse | `tool.execute.before` | Blocks invalid reconstructable artifact candidates by throwing an error |
 | PostToolUse | `tool.execute.after` | Fires after every tool execution |
-| Compaction context | `experimental.session.compacting` | Appends bounded Agentera state from `agentera hej --format json` |
+| Compaction context | `experimental.session.compacting` | Appends bounded Agentera state from `agentera prime --format json` |
 
 OpenCode plugins subscribe to session events by exporting one generic `event` hook:
 
@@ -233,7 +230,7 @@ export const AgenteraPlugin = async ({ project, client, $, directory, worktree }
       // PostToolUse equivalent: report validation warnings
     },
     "chat.message": async (input, output) => {
-      // Exact bare hej prompt rewrite.
+      // Exact bare /agentera prompt rewrite.
     },
     "experimental.session.compacting": async (input, output) => {
       // Append bounded Agentera CLI state context.
@@ -248,17 +245,17 @@ Install-root semantics used by the plugin are contract-bound to `scripts/install
 
 OpenCode runtime facts are owned by the RuntimeAdapter registry at `references/adapters/runtime-adapter-registry.yaml` and loaded through `scripts/runtime_adapter_registry.py`. This reference describes the OpenCode adapter behavior, but registry-owned values such as event support, artifact-validation claims, config targets, diagnostic labels, and documentation claims must be validated against the registry rather than changed here as an independent source. The install-root Module must keep package manifest registry and package metadata consolidation work outside its scope.
 
-Registry parity claims: Blocks invalid reconstructable artifact candidates. Routes exact bare text hej through chat.message to the bundled Agentera dashboard path.
+Registry parity claims: Blocks invalid reconstructable artifact candidates. Routes exact bare text /agentera through chat.message to the bundled Agentera dashboard path.
 
 1. **SessionStart** (`event.type === "session.created"`): Observe session creation through the event hook. Session-start context preload remains deferred because no supported model-context injection path is verified for this adapter. Do not attach dead preload code to event observation alone.
 
 2. **PreToolUse** (`tool.execute.before`): When `write` or `edit` args expose a path plus candidate content or exact replacement evidence, validate the candidate through `hooks/validate_artifact.py`. Throw an error to block invalid artifact content before mutation. Sparse payloads and `apply_patch` `patchText` without reconstructed full content are allowed rather than guessed.
 
-3. **Exact bare `hej`** (`chat.message`): If the incoming user message has exactly one meaningful text part and that text is exactly `hej` after removing only OpenCode's CLI-added single trailing newline transport artifact, replace that text with an Agentera routing prompt that preserves the original-message fact and instructs the agent to run `agentera hej` instead of replying with a generic greeting. This is intentionally not a broad greeting detector.
+3. **Exact bare `/agentera`** (`chat.message`): If the incoming user message has exactly one meaningful text part and that text is exactly `/agentera` after removing only OpenCode's CLI-added single trailing newline transport artifact, replace that text with an Agentera routing prompt that preserves the original-message fact and instructs the agent to run `agentera prime` instead of replying with a generic greeting. This is intentionally not a broad greeting detector.
 
 4. **PostToolUse** (`tool.execute.after`): After `write` or `edit`, run the shared artifact validator for warnings. This is advisory because the mutation already happened.
 
-5. **Compaction context** (`experimental.session.compacting`): Before OpenCode compacts a session, append a bounded Agentera project-state summary generated by `agentera hej --format json`. This hook is CLI-first by design and does not read `.agentera` artifacts directly.
+5. **Compaction context** (`experimental.session.compacting`): Before OpenCode compacts a session, append a bounded Agentera project-state summary generated by `agentera prime --format json`. This hook is CLI-first by design and does not read `.agentera` artifacts directly.
 
 6. **Stop** (`event.type === "session.idle"`): Append a runtime-local bookmark entry under the Agentera data directory capturing artifact changes for next-session continuity.
 
@@ -276,9 +273,11 @@ This keeps the plugin thin while matching the packaged Python script contract.
 
 Profilera mines five canonical record types from host session data. This section maps each record type to OpenCode's data sources.
 
-### memory_entry (Claude Code runtime extension)
+### memory_entry (retired historical-import extension)
 
-**Agentera contract**: This record type is a Claude Code runtime extension, not part of the portable corpus. Claude Code's memory files are emitted as instruction_document records with `doc_type: "claude_memory"`.
+**Agentera contract**: This retired-source record type is not part of the
+portable active corpus. It may appear only in explicitly consented historical
+imports and remains excluded from default analytics.
 
 **OpenCode source**: OpenCode does not have a built-in memory system. The memory_entry extension does not apply.
 
@@ -315,7 +314,7 @@ Profilera mines five canonical record types from host session data. This section
 }
 ```
 
-**Gap**: None. OpenCode has richer instruction document support than Claude Code (remote URLs, glob patterns, explicit config).
+**Gap**: None. OpenCode supports remote URLs, glob patterns, and explicit config.
 
 ### history_prompt
 
@@ -336,7 +335,7 @@ The SDK exposes message history with timestamps, project context, and session me
 1. Enumerate sessions via `opencode session list --format json`
 2. Export each session via `opencode export [sessionID]`
 3. Filter for user-originated messages from the export JSON
-4. Apply decision-pattern regex (same patterns profilera uses for Claude Code history) to classify prompts as `"decision"`, `"correction"`, or `"question"`
+4. Apply the shared decision-pattern classifier to label prompts as `"decision"`, `"correction"`, or `"question"`
 5. Produce history_prompt records with session and project metadata
 
 ```python
@@ -400,13 +399,13 @@ The SDK approach remains preferable for programmatic integration (e.g., as an Op
 
 **OpenCode source**: OpenCode projects are standard filesystem directories. Config files are accessible via direct file reads.
 
-**Adapter extraction**: Identical to Claude Code extraction. Scan the project root for known config types:
+**Adapter extraction**: Scan the project root for known config types:
 
 - `package.json`, `go.mod`, `Cargo.toml`, `pyproject.toml`
 - `.golangci.yml`, `tsconfig.json`, `ruff.toml`
 - `Makefile`, `magefile.go`, `justfile`, `taskfile.yaml`
 
-No runtime-specific adaptation needed. The adapter uses the same config type list and extraction logic as the Claude Code adapter.
+No runtime-specific adaptation is needed; the config scan is runtime-agnostic.
 
 ```python
 {
@@ -436,9 +435,9 @@ Summary of which corpus families the OpenCode adapter can produce:
 | Crystallized decisions | instruction_document | Yes | AGENTS.md files (global and project), instructions config |
 | Decision history | history_prompt | Yes (CLI) | `opencode session list --format json` enumerates sessions; `opencode export [sessionID]` provides full message JSON for prompt extraction |
 | Conversation exchanges | conversation_turn | Yes (CLI) | `opencode export [sessionID]` provides full session JSON with paired user-assistant turns |
-| Config patterns | project_config_signal | Yes | Direct filesystem scan, identical to Claude Code |
+| Config patterns | project_config_signal | Yes | Direct filesystem scan |
 
-**Initial port profilera mode**: Full. All four source families are available via included OpenCode functionality. Crystallized decisions come from AGENTS.md files and instructions config. Decision history and conversation exchanges come from `opencode export` CLI output. Config patterns come from direct filesystem scan.
+**Initial port profile mode**: Full. All four source families are available via included OpenCode functionality. Crystallized decisions come from AGENTS.md files and instructions config. Decision history and conversation exchanges come from `opencode export` CLI output. Config patterns come from direct filesystem scan.
 
 ---
 
@@ -510,8 +509,8 @@ OpenCode repair is ownership-gated. Managed slash commands include `agentera_man
 
 | Gap | Impact | Mitigation |
 |-----|--------|------------|
-| Sub-agent dispatch lacks built-in worktree isolation | realisera/orkestrera run in same working tree | Strategy A (Task tool) for initial port; manual `git worktree` commands for full isolation |
-| `opencode run --format json` event schema differs from `claude -p --output-format json` | Eval runner needs a separate JSON event parser | Implement an OpenCode-specific parser module in eval_skills.py; event stream structure is straightforward |
+| Sub-agent dispatch lacks built-in worktree isolation | build/orchestrate run in same working tree | Strategy A (Task tool) for initial port; manual `git worktree` commands for full isolation |
+| `opencode run --format json` uses an OpenCode-specific event schema | Eval runner needs a dedicated parser | Use the OpenCode parser module in eval_skills.py; event stream structure is straightforward |
 | Session history requires JSON event schema mapping | history_prompt and conversation_turn need export parser | `opencode export [sessionID]` provides full session JSON; adapter parses the export output |
 | Python scripts require Python runtime | Hook plugin calls Python via shell | Python is already a prerequisite for agentera scripts |
 
@@ -523,7 +522,7 @@ To verify the adapter is sufficient, check each acceptance criterion from PLAN.m
 
 1. **Each of the six host capabilities is mapped**: Yes. Skill discovery, artifact resolution, and profile path have direct OpenCode equivalents with no gaps. Sub-agent dispatch maps to the Task tool (with manual git worktree as an alternative). Eval mechanism maps to `opencode run --format json`. Hook lifecycle maps to the plugin event system.
 
-2. **Session Corpus Contract is mapped per record type**: Yes. All four portable record types are mapped. instruction_document and project_config_signal have immediate extraction paths. history_prompt and conversation_turn are extractable via `opencode export` CLI output. The memory_entry Claude Code extension does not apply to OpenCode.
+2. **Session Corpus Contract is mapped per record type**: Yes. All four portable record types are mapped. instruction_document and project_config_signal have immediate extraction paths. history_prompt and conversation_turn are extractable via `opencode export` CLI output. The retired historical-import extension does not apply to OpenCode.
 
 3. **A developer can implement OpenCode support without reading skill entry file source**: Yes. This document specifies the adapter mapping, installation steps, extraction logic, and remaining gaps independently.
 

package/bundle/references/adapters/package-manifest-interface-model.yaml

@@ -5,7 +5,7 @@ intent: >
   Define the package and release-surface access Interface before introducing an
   executable registry Module. This model records ownership boundaries and the
   validation contract for future PackageManifest records. The executable registry
-  at references/adapters/package-registry.yaml and its loader scripts/package_registry.py
+  at references/adapters/package-registry.yaml and its TypeScript PackageRegistry loader
   now implement this interface; this model remains the authoritative design contract.
 record:
   required_groups:
@@ -115,7 +115,7 @@ ownership:
       explicit future ADR changes persisted authority.
     future_authority_change_requires: explicit ADR
   install_root_delegated:
-    owner: scripts/install_root.py
+    owner: packages/cli/src/state/installRoot.ts
     forbidden_in_package_manifest:
       - AGENTERA_HOME precedence
       - default app home
@@ -124,7 +124,7 @@ ownership:
       - unmanaged classification
       - app-home diagnostics
   runtime_adapter_delegated:
-    owner: scripts/runtime_adapter_registry.py
+    owner: packages/cli/src/registries/runtimeAdapterRegistry.ts
     forbidden_in_package_manifest:
       - runtime discovery
       - lifecycle events
@@ -141,10 +141,11 @@ command_safety:
     - install-agentera-skill
   approved_runtimes:
     - all
-    - claude
     - opencode
+    - codex
+    - cursor
+    - copilot
   approved_runtime_agents:
-    - claude-code
     - opencode
   approved_package_tool:
     executable: npx
@@ -176,9 +177,12 @@ sample_manifest:
       - id: registry
         path: registry.json
         selector: skills[0].version
-      - id: python-project
-        path: pyproject.toml
-        selector: project.version
+      - id: cli-package
+        path: packages/cli/package.json
+        selector: version
+      - id: cli-suite-marker
+        path: packages/cli/package.json
+        selector: agentera.suiteVersion
       - id: copilot-root
         path: plugin.json
         selector: version
@@ -188,15 +192,15 @@ sample_manifest:
       - id: codex-plugin
         path: .codex-plugin/plugin.json
         selector: version
-      - id: claude-marketplace-metadata
-        path: .claude-plugin/marketplace.json
-        selector: metadata.version
-      - id: claude-marketplace-plugins
-        path: .claude-plugin/marketplace.json
-        selector: plugins[*].version
+      - id: cursor-plugin
+        path: .cursor-plugin/plugin.json
+        selector: version
       - id: opencode-plugin-marker
         path: .opencode/plugins/agentera.js
         selector: AGENTERA_VERSION
+      - id: skill-frontmatter
+        path: skills/agentera/SKILL.md
+        selector: frontmatter.version
     excluded_runtime_manifests:
       - .opencode/package.json
   runtime_package_manifests:
@@ -213,28 +217,27 @@ sample_manifest:
         path: .codex-plugin/plugin.json
         version_bearing: true
         package_shape: suite-bundle
-      - runtime: claude
-        path: .claude-plugin/marketplace.json
+      - runtime: cursor
+        path: .cursor-plugin/plugin.json
         version_bearing: true
         package_shape: suite-bundle
       - runtime: opencode
         path: .opencode/package.json
         version_bearing: false
-        package_shape: suite-bundle
+        package_shape: npm-loadable-plugin
     shared_paths_policy: validate minimum shared paths separately from full bundle includes
   bundle_surfaces:
     directories:
       - skills
-      - scripts
-      - hooks
       - references
       - agents
-      - .agents/plugins
-      - .codex-plugin
-      - .claude-plugin
+      - hooks
       - .github/hooks
-      - .github/plugin
+      - .codex-plugin
+      - .cursor-plugin
+      - .cursor/agents
       - .opencode/commands
+      - .opencode/agents
       - .opencode/plugins
     files:
       - README.md
@@ -242,11 +245,11 @@ sample_manifest:
       - CHANGELOG.md
       - DESIGN.md
       - LICENSE
-      - pyproject.toml
-      - uv.lock
       - registry.json
       - plugin.json
+      - .github/plugin/plugin.json
       - .opencode/package.json
+      - .cursor/hooks.json
     skip_parts:
       - __pycache__
       - .pytest_cache
@@ -278,21 +281,6 @@ sample_manifest:
           - -g
           - -y
         skipped_without_update_packages_message: legacy skill removal skipped; pass --update-packages to run
-      - runtime: claude
-        action: install-agentera-skill
-        phase: runtime-install
-        argv:
-          - npx
-          - skills
-          - add
-          - jgabor/agentera
-          - -g
-          - -a
-          - claude-code
-          - --skill
-          - agentera
-          - -y
-        skipped_without_update_packages_message: external package update skipped; pass --update-packages to run
       - runtime: opencode
         action: install-agentera-skill
         phase: runtime-install
@@ -318,17 +306,18 @@ sample_manifest:
   docs_targets:
     version_files_source: .agentera/docs.yaml conventions.version_files
     version_files:
-      - pyproject.toml
-      - uv.lock
+      - packages/cli/package.json
       - plugin.json
       - .github/plugin/plugin.json
       - .codex-plugin/plugin.json
-      - .claude-plugin/marketplace.json
+      - .cursor-plugin/plugin.json
       - .opencode/plugins/agentera.js
+      - skills/agentera/SKILL.md
       - registry.json
     index_targets:
       - references/adapters/package-surface-characterization.md
       - references/adapters/package-manifest-interface-model.yaml
+      - references/adapters/package-registry.yaml
     excluded_version_files:
       - .opencode/package.json
   release_policy:

package/bundle/references/adapters/package-registry.yaml

@@ -17,9 +17,12 @@ records:
         - id: registry
           path: registry.json
           selector: skills[0].version
-        - id: python-project
-          path: pyproject.toml
-          selector: project.version
+        - id: cli-package
+          path: packages/cli/package.json
+          selector: version
+        - id: cli-suite-marker
+          path: packages/cli/package.json
+          selector: agentera.suiteVersion
         - id: copilot-root
           path: plugin.json
           selector: version
@@ -29,15 +32,15 @@ records:
         - id: codex-plugin
           path: .codex-plugin/plugin.json
           selector: version
-        - id: claude-marketplace-metadata
-          path: .claude-plugin/marketplace.json
-          selector: metadata.version
-        - id: claude-marketplace-plugins
-          path: .claude-plugin/marketplace.json
-          selector: plugins[*].version
+        - id: cursor-plugin
+          path: .cursor-plugin/plugin.json
+          selector: version
         - id: opencode-plugin-marker
           path: .opencode/plugins/agentera.js
           selector: AGENTERA_VERSION
+        - id: skill-frontmatter
+          path: skills/agentera/SKILL.md
+          selector: frontmatter.version
       excluded_runtime_manifests:
         - .opencode/package.json
     runtime_package_manifests:
@@ -62,11 +65,6 @@ records:
           path: .cursor-plugin/plugin.json
           version_bearing: true
           package_shape: suite-bundle
-        - id: claude-marketplace-manifest
-          runtime: claude
-          path: .claude-plugin/marketplace.json
-          version_bearing: true
-          package_shape: suite-bundle
         - id: opencode-package-manifest
           runtime: opencode
           path: .opencode/package.json
@@ -76,9 +74,6 @@ records:
         - id: skills
           path: skills
           kind: dir
-        - id: scripts
-          path: scripts
-          kind: dir
         - id: hooks
           path: hooks
           kind: dir
@@ -99,28 +94,20 @@ records:
       directories:
         - id: skills
           path: skills
-        - id: scripts
-          path: scripts
-        - id: hooks
-          path: hooks
         - id: references
           path: references
         - id: agents
           path: agents
-        - id: agents-plugins
-          path: .agents/plugins
+        - id: hooks
+          path: hooks
+        - id: github-hooks
+          path: .github/hooks
         - id: codex-plugin
           path: .codex-plugin
         - id: cursor-plugin
           path: .cursor-plugin
         - id: cursor-agents
           path: .cursor/agents
-        - id: claude-plugin
-          path: .claude-plugin
-        - id: github-hooks
-          path: .github/hooks
-        - id: github-plugin
-          path: .github/plugin
         - id: opencode-commands
           path: .opencode/commands
         - id: opencode-agents
@@ -138,14 +125,12 @@ records:
           path: DESIGN.md
         - id: license
           path: LICENSE
-        - id: python-project
-          path: pyproject.toml
-        - id: python-lock
-          path: uv.lock
         - id: registry
           path: registry.json
         - id: copilot-root-manifest
           path: plugin.json
+        - id: copilot-repository-manifest
+          path: .github/plugin/plugin.json
         - id: opencode-package-manifest
           path: .opencode/package.json
         - id: cursor-hooks
@@ -182,22 +167,6 @@ records:
             - -g
             - -y
           skipped_without_update_packages_message: legacy skill removal skipped; pass --update-packages to run
-        - id: install-agentera-skill-claude
-          runtime: claude
-          action: install-agentera-skill
-          phase: runtime-install
-          argv:
-            - npx
-            - skills
-            - add
-            - jgabor/agentera
-            - -g
-            - -a
-            - claude-code
-            - --skill
-            - agentera
-            - -y
-          skipped_without_update_packages_message: external package update skipped; pass --update-packages to run
         - id: install-agentera-skill-opencode
           runtime: opencode
           action: install-agentera-skill
@@ -224,16 +193,13 @@ records:
     docs_targets:
       version_files_source: .agentera/docs.yaml conventions.version_files
       version_files:
-        - pyproject.toml
-        - uv.lock
+        - packages/cli/package.json
         - plugin.json
         - .github/plugin/plugin.json
         - .codex-plugin/plugin.json
         - .cursor-plugin/plugin.json
-        - .claude-plugin/marketplace.json
         - .opencode/plugins/agentera.js
         - skills/agentera/SKILL.md
-        - skills/hej/SKILL.md
         - registry.json
       index_targets:
         - references/adapters/package-surface-characterization.md