agentera 3.0.0-dev.25 → 3.0.0-dev.26

package/bundle/references/cli/bundle-skill-vocabulary.yaml

@@ -12,7 +12,7 @@ scope:
   includes:
     - Current user-facing and diagnostic prose that names Agentera app files.
     - Package and distribution metadata that names the suite package shape.
-    - Runtime hook packaging prose and metadata for plugin-shipped hooks.
+    - Historical classification of retired runtime hook packaging prose and metadata.
     - Historical references to the removed `bundle-status` command.
     - Current references to `skills/agentera/SKILL.md`.
   excludes:
@@ -52,8 +52,7 @@ canonical_concepts:
   suite_package:
     definition: >-
       The packaged distribution shape that ships the Agentera skill, CLI,
-      scripts, hooks, references, runtime descriptors, and package metadata as
-      one suite artifact.
+      references, and package metadata as one suite artifact.
     allowed_terms:
       - suite package
       - Agentera package
@@ -63,10 +62,10 @@ canonical_concepts:
       - bundle
       - single-bundle model
       - bundled suite
-  plugin_shipped_hooks:
+  retired_plugin_shipped_hooks:
     definition: >-
-      Hooks that are shipped by active runtime plugin package surfaces such as
-      Codex, OpenCode, Cursor, or GitHub plugin metadata.
+      Historical term for hooks shipped by retired v2 host-native packages.
+      Current Agentera packages ship no host-native hooks or plugin metadata.
     allowed_terms:
       - plugin-shipped hooks
       - plugin-bundled hooks
@@ -155,6 +154,7 @@ focused_scan:
   allowed_classifications:
     - canonical
     - compatibility-preserved
+    - migration-only
     - historical
     - fixture-only
     - generic
@@ -162,7 +162,6 @@ focused_scan:
     - AGENTS.md
     - README.md
     - references/cli/vocabulary.md
-    - references/adapters/opencode.md
     - (retired maintainer measurement tool)
     - skills/agentera/SKILL.md
     - skills/agentera/capabilities/hej/schemas/triggers.yaml
@@ -193,16 +192,8 @@ focused_scan:
       reason: Vocabulary docs use concrete skill entry file paths and source-index references.
     - path: ^references/cli/vocabulary\.md$
       text: --only bundle|\.agentera-bundle\.json|bundle\.status|activeBundleRoot|bundle-status
-      classification: compatibility-preserved
-      reason: Vocabulary docs preserve shipped compatibility identifiers and removed command references.
-    - path: ^references/adapters/opencode\.md$
-      text: bundled Agentera dashboard path|bundled `agentera` skill|bundled Agentera
-      classification: generic
-      reason: OpenCode adapter prose uses bundled in the ordinary package-shipping sense, not as Agentera app-file terminology.
-    - path: ^references/adapters/opencode\.md$
-      text: SKILL\.md files|<name>/SKILL\.md|skills/\*/SKILL\.md|skills/agentera/SKILL\.md
-      classification: compatibility-preserved
-      reason: OpenCode adapter prose documents concrete runtime skill entry file paths.
+      classification: migration-only
+      reason: Vocabulary docs classify retired v2 identifiers for migration evidence and released history only.
     - path: ^scripts/measure_token_payload\.py$
       text: 'v1 baseline: all 12 v1 SKILL\.md files|skill_root / "SKILL\.md"'
       classification: historical
@@ -223,7 +214,7 @@ focused_scan:
       text: .
       classification: fixture-only
       reason: Runtime adapter tests use fixture paths, expected diagnostics, package metadata, and compatibility test class names.
-compatibility_identifiers:
+migration_identifiers:
   bundle:
     - .agentera-bundle.json
     - bundle.status

package/bundle/references/cli/capability-tool-classification.yaml

@@ -1,7 +1,7 @@
 # Capability tool classification authority
 # Defines which tools each capability subagent needs to fulfill its purpose.
-# Consumed by: descriptor validation, regression tests, runtime adapter docs.
-# Source of truth for per-capability tool requirements across all runtimes.
+# Consumed by shared-skill worker metadata validation and regression tests.
+# Hosts provide and enforce their own tool surfaces; Agentera ships no native adapter.
 
 classification:
   mutation_capable:

package/bundle/references/cli/prime-consumer-compatibility.yaml

@@ -22,34 +22,6 @@ exclusions:
 # Tasks 2-4 MUST keep every consumed field resolvable from a declared surface.
 # ---------------------------------------------------------------------------
 consumers:
-  - id: opencode_compaction_context
-    startup_command: agentera prime --format json
-    evidence: .opencode/plugins/agentera.js:502-517
-    consumed_fields:
-      - mode
-      - profile.status
-      - profile
-      - health.grade
-      - health.trajectory
-      - health
-      - issues
-      - plan
-      - plan.exists
-      - plan.status
-      - plan.title
-      - plan.first_pending.name
-      - progress
-      - progress.latest
-      - progress.latest.id
-      - progress.latest.what
-      - progress.latest.next
-      - next_action
-      - next_action.object
-      - attention
-    notes: >-
-      Reads the deprecated `issues` alias instead of `todo` (lines 480-486).
-      This is the only documented runtime consumer of the deprecated alias, so
-      `issues` is transition-required until the stable cut.
   - id: status_dashboard_renderer
     startup_command: agentera prime --context status --format json
     evidence: skills/agentera/SKILL.md:68
@@ -75,43 +47,35 @@ consumers:
       through the named command in the status context, never by raw artifact read.
   - id: capability_context_dispatcher
     startup_command: agentera prime --context <capability> --format json
-    evidence: .opencode/agents/agentera.md:12
-    alt_evidence: .cursor/agents/agentera.md:7
-    runtime_evidence: packages/cli/src/cli/capabilityContext/startup.ts:250
+    evidence: packages/cli/src/cli/capabilityContext/startup.ts:250
+    skill_evidence: skills/agentera/SKILL.md:68-103
     consumed_fields:
       - command
       - status
       - capability_context
       - capability_context.instructions
     notes: >-
-      Per-capability startup capsule. The runtime emits
+      Per-capability startup capsule. The CLI emits
       `capability_context.instructions` (the capability's instructions module),
-      not `prose`. The agent descriptors (.opencode/agents/agentera.md and
-      .cursor/agents/agentera.md) formerly documented a `prose` field that the
-      v3 runtime never emits; they were reconciled to
-      `capability_context.instructions` to match
-      packages/cli/src/cli/capabilityContext/startup.ts:250 and
-      skills/agentera/SKILL.md (lines 68, 103), which already documented
-      `capability_context.instructions`. This consumer governs the `--context`
+      not `prose`; the shared skill consumes that field directly. This consumer
+      governs the `--context`
       surface, not the bare briefing; the bounded decision brief (Task 3) must
       not break the `--context` capsule. Executable fixture proves the emitted
       payload contains `capability_context.instructions` and NOT `prose`.
   - id: packaging_distribution_gate
     startup_command: agentera prime --format json
-    evidence: docs/packaging/v3-packaging.md:130-139
+    evidence: packages/cli/test/packaging/packageVerification.test.ts
     consumed_fields:
       - command
       - status
       - app_home
       - app_home.source
     notes: >-
-      Distribution/packaging verification matrix consumer. The Prime gate
-      (line 136) asserts `command: "prime"`, `status: "ok"`, and that
-      `app_home.source` reflects resolution; the Bun smoke gate (line 139)
-      asserts the published single binary runs `agentera prime --format json`
-      and exits 0 with the same JSON shape. Tasks 2-4 must keep the published
-      prime JSON shape stable so packaged binaries and fresh-cache npx gates
-      keep passing. Not a dashboard consumer; does not read state families.
+      Distribution-boundary consumer. The one extracted-package conjunction
+      asserts `command: "prime"`, `status: "ok"`, and that `app_home.source`
+      reflects resolution after the packed upgrade path. Tasks 2-4 must keep
+      the published prime JSON shape stable so fresh-cache npx gates keep
+      passing. Not a dashboard consumer; does not read state families.
   - id: json_output_surface_manifest
     startup_command: agentera prime [ --dashboard | --context <capability> ] --format json
     evidence: scripts/json_output_surface_manifest.yaml:47-161
@@ -136,9 +100,8 @@ consumers:
       from the retired `prose` field to the v3 `capability_context.instructions`
       field after the second Task 1 evaluation found this guidance was the one
       remaining live consumer contradiction. Runtime authority
-      (capabilityContext/startup.ts:250), SKILL.md (lines 68, 103), and the
-      agent descriptors (.opencode/agents/agentera.md, .cursor/agents/agentera.md)
-      all already used `capability_context.instructions`. The
+      (capabilityContext/startup.ts:250) and SKILL.md (lines 68, 103) both use
+      `capability_context.instructions`. The
       `documentation_drift` guard below prevents this guidance from drifting
       back to `prose`; it does not execute a prime command.
 
@@ -154,7 +117,7 @@ consumers:
 # and kept until the 3.0.0 stable cut; it is excluded only from the canonical
 # advertised `source_contract.fields` set (one canonical representation — `todo`
 # is canonical). It is NOT selector-only: it ships in the default bare briefing
-# for the opencode_compaction_context consumer that still reads it.
+# only as bounded CLI transition compatibility.
 # ---------------------------------------------------------------------------
 field_authorities:
   - id: orientation_output_emission
@@ -194,7 +157,7 @@ field_authorities:
     notes: >-
       Reconciled by Task 2: both hej.fields lists now mirror the canonical
       27-field set (retired `bundle` dropped; app, app_home, project_integration,
-      runtime_lifecycle, and history added).
+      shared_skill, project_integration, and history added).
 
 # ---------------------------------------------------------------------------
 # Field classification. Every field proposed for omission is classified with a
@@ -264,8 +227,8 @@ field_classification:
   - field: project_integration
     status: required
     rationale: >-
-      Status routing consumes the recommendation and aggregate lifecycle state;
-      the bounded brief retains the routing projection even when no upgrade
+      Status routing consumes the app and project-state recommendation; the
+      bounded brief retains that routing projection even when no upgrade
       recommendation is active.
   - field: attention
     status: required
@@ -276,11 +239,9 @@ field_classification:
       Startup history. The bounded brief retains counts and named retrieval
       pointers while omitting entry detail; explicit state list/get commands
       recover the current history.
-  - field: runtime_lifecycle
-    status: conditional
-    rationale: >-
-      Diagnostic. Plan constraint: unrelated lifecycle diagnostics are omitted
-      unless status declares them necessary for the current brief (Task 4).
+  - field: shared_skill
+    status: required
+    rationale: Canonical shared-skill path and read-only diagnosis.
   - field: decision_attention
     status: conditional
     rationale: Null unless returned decisions require satisfaction review.
@@ -326,10 +287,9 @@ field_classification:
   - field: issues
     status: deprecated_transition
     rationale: >-
-      Deprecated alias of `todo`. opencode_compaction_context reads it
-      (.opencode/plugins/agentera.js:480-486). Removable at the 3.0.0 stable cut
-      only; tracked in TODO.md [fix:3.0.0]. Tasks 2-4 must keep the alias and
-      the stderr deprecation warning.
+      Deprecated alias of `todo`. No current consumer in this inventory reads
+      it; the alias remains only as bounded CLI transition compatibility until
+      its owning contract removes it.
   - field: bundle
     status: retired
     rationale: >-
@@ -369,9 +329,9 @@ public_output_policy:
     separate bare-prime call or raw artifact read. The `--context` capsule
     contract (command/status/capability_context) is unchanged. The status
     capsule puts the canonical Task 3 projection at
-    `capability_context.context.status_context`; unrelated per-runtime
-    lifecycle diagnostics are omitted while aggregate release-blocking state
-    remains available for routing.
+    `capability_context.context.status_context`; app/project integration and
+    shared-skill diagnosis remain available for routing while unrelated detail
+    is omitted.
   status_context_contract:
     startup_command: agentera prime --context status --format json
     state_path: capability_context.context.status_context
@@ -474,7 +434,7 @@ default_emission_omission_contract:
       only; diagnostics measured separately on stderr.
     retained_top_level_fields: >-
       Every field in never_omitted plus command, status, app_home, app, mode,
-      history, runtime_lifecycle, project_integration, decision_attention, and
+      history, shared_skill, project_integration, decision_attention, and
       the bespoke context pointers (kept null on the bare surface). The
       conditional v1_migration/docs/objective are retained when active (see
       omitted_when_default). issues ships as the deprecated transition alias.
@@ -491,9 +451,6 @@ default_emission_omission_contract:
       - field: history.{progress,decisions,health}.entries
         reason: startup_history_entries
         recovery: agentera state <artifact> list --limit 20 --format json
-      - field: runtime_lifecycle.runtimes
-        reason: lifecycle_runtime_detail
-        recovery: agentera upgrade --dry-run --format json
       - field: project_integration.phases
         reason: phase_blockers
         recovery: agentera doctor --format json
@@ -546,7 +503,7 @@ declared_available_fields:
   - project_integration
   - attention
   - history
-  - runtime_lifecycle
+  - shared_skill
   - decision_attention
   - next_action
   - orchestration_context
@@ -560,33 +517,6 @@ declared_available_fields:
   - issues
 fixtures:
   pass:
-    - consumer: opencode_compaction_context
-      executable_command: agentera prime --format json
-      depends_on:
-        - mode
-        - profile
-        - health
-        - issues
-        - plan
-        - progress
-        - next_action
-        - attention
-      nested_fields:
-        - profile.status
-        - next_action.object
-      conditional_nested_fields:
-        - path: health.grade
-          present_when: health.exists
-        - path: health.trajectory
-          present_when: health.exists
-        - path: plan.first_pending.name
-          present_when: plan.first_pending
-        - path: progress.latest.id
-          present_when: progress.latest
-        - path: progress.latest.what
-          present_when: progress.latest
-        - path: progress.latest.next
-          present_when: progress.latest
     - consumer: status_dashboard_renderer
       executable_command: agentera prime --context status --format json
       depends_on:

package/bundle/references/cli/routing-execution-vocabulary.yaml

@@ -2,7 +2,7 @@ schema_version: agentera.routing_execution_vocabulary.v1
 status: active_authority
 purpose: >-
   Define routing and execution vocabulary boundaries so request routing, next-action
-  recommendation, Orkestrera task assignment, worker launch, runtime support, and
+  recommendation, Orkestrera task assignment, worker launch, host-provided worker support, and
   worker-safety Git commits stay distinct without forcing mechanical renames of
   shipped compatibility identifiers or historical evidence.
 sources:
@@ -14,7 +14,7 @@ scope:
     - Hej and capability next-action recommendations that wait for confirmation.
     - Orkestrera assignment of approved plan work to worker capabilities.
     - Realisera and Optimera launch of runtime workers.
-    - Runtime subagent support mechanisms and compatibility identifiers.
+    - Host-provided worker mechanisms without Agentera-owned runtime integration.
     - Git commit safety language before spawning isolated worker execution.
   excludes:
     - Broad current-prose replacement across capability instructions.
@@ -96,8 +96,8 @@ canonical_concepts:
       - launch substrate
   runtime_subagent_support:
     definition: >-
-      Runtime-specific support that lets Agentera use a host's subagent surface,
-      descriptor model, or equivalent worker mechanism.
+      A worker mechanism supplied by the coding host. Agentera may request a
+      worker but ships no host-specific descriptor, adapter, plugin, or hook.
     preferred_terms:
       - subagent mechanism
       - runtime subagent mechanism
@@ -167,7 +167,6 @@ focused_scan:
   paths:
     - AGENTS.md
     - references/cli/vocabulary.md
-    - references/adapters/runtime-feature-parity.md
     - (retired maintainer measurement tool)
     - skills/agentera/SKILL.md
     - skills/agentera/references/contract.md
@@ -198,10 +197,6 @@ focused_scan:
       text: subagent_dispatch|fixture names
       classification: compatibility-preserved
       reason: Routing vocabulary docs preserve explicit compatibility identifiers and fixture terminology.
-    - path: ^references/adapters/runtime-feature-parity\.md$
-      text: prompt rewrite router|Subagent Dispatch|Dispatch surface|Capability dispatch|subagent dispatch
-      classification: compatibility-preserved
-      reason: Runtime parity docs name concrete host and compatibility surfaces around descriptor dispatch support.
     - path: ^scripts/measure_token_payload\.py$
       text: static dispatch payload|Agentera router
       classification: historical
@@ -238,20 +233,7 @@ focused_scan:
       text: '`dispatch`: invoke another capability autonomously|`chain`: dispatch multiple capabilities autonomously|In orkestrera only, `dispatch` and `chain` are autonomous'
       classification: fixture-only
       reason: Skill dispatch contract tests assert the canonical Orkestrera dispatch reservation.
-    - path: ^tests/test_runtime_adapters\.py$
-      text: opencode bare-hej router|routeBareHejMessage|subagent_dispatch
-      classification: fixture-only
-      reason: Runtime adapter tests assert plugin routing helpers and compatibility identifier fixtures.
-    - path: ^tests/test_runtime_adapter_registry\.py$
-      text: subagent_dispatch
-      classification: code-identifier
-      reason: Runtime adapter registry tests preserve the shipped subagent_dispatch field name.
 compatibility_identifiers:
-  runtime_adapter:
-    - subagent_dispatch
-    - RuntimeAdapter.subagent_dispatch
-    - subagent_dispatch.supported
-    - subagent_dispatch.group
   route_aliases:
     - ROUTE_ALIASES
     - primary_aliases

package/bundle/references/cli/upgrade-repair-wording.md

@@ -1,34 +1,14 @@
-# Upgrade repair wording — v3 cursor agent skip
+# Upgrade repair wording
 
-When `agentera upgrade --runtime cursor` (or a full runtime phase) runs against a
-**project** directory, managed `.cursor/agents/*.md` descriptors are normally
-copied from the active app-home bundle. On a **v3 capability-surface** project
-that signal is suppressed so D65 `prime --context` bodies are not regressed to
-the retired `instructions.md` read path.
+## Runtime migration boundary
 
-## v3 surface probe
+Recognized v2-to-v3 project migration rewires existing Agentera v2 hook
+configurations to the npm CLI and retires only proven Agentera-owned legacy hook
+files. It does not copy or refresh plugins, agents, commands, or skill links.
 
-Treat the project as v3 when **all twelve** instruction modules exist:
-
-`packages/cli/src/capabilities/<name>/instructions.ts`
-
-for every routed capability (`status`, `vision`, `discuss`, `research`,
-`plan`, `build`, `optimize`, `audit`, `document`, `profile`,
-`design`, `orchestrate`).
-
-If any module is missing, preserve v2 upgrade behavior (copy/refresh managed
-agents from the bundle).
-
-## Orchestrator behavior
-
-| Orchestrator | Branch | Behavior |
-| --- | --- | --- |
-| TypeScript (`packages/cli/src/upgrade/runtimeMigration.ts`) | `feat/v3` | On v3 probe match, emit `skipped` `copy-agent` items for in-tree `.cursor/agents/` instead of copying from the bundle. |
-| Python (`scripts/agentera_upgrade.py`) | `main` (stable) | Same probe before the cursor agent copy loop; do not schedule `copy-agent` targets under the project when the probe matches. Backport per the both-branch pattern. |
-
-## User-visible skip message
-
-`v3 capability instruction modules present; in-tree .cursor/agents/ uses prime --context and is not overwritten`
+Runtime lifecycle installation and remediation are separate. Apply requires an
+explicit `--runtime` selector and `--yes`; declared ownership, collision
+remediation, and convergent retry behavior remain governed by `UPGRADE.md`.
 
 ## v2 install track successor surfacing (#32)
 

package/bundle/references/cli/v3-handoff-manifest.schema.yaml

@@ -47,9 +47,8 @@ required_top_level_fields:
     items:
       type: string
     description: >-
-      Runtime adapter ids (references/adapters/runtime-adapter-registry.yaml)
-      that the v2 install actively uses — runtimes whose setup doctor status is
-      pass at write time.
+      Historical runtime adapter ids that the v2 install actively used at handoff.
+      The v3 reader treats these as migration evidence, not active runtime inventory.
 user_data_inventory_catalog:
   - id: benchmarks
     relative_path: benchmarks

package/bundle/references/cli/vocabulary.md

@@ -51,11 +51,11 @@ examples. Diagnostics should state object, state, cause, and fix.
 
 | Term                       | Definition                                                                                                                                                                                                                                                                          | Common sources                                                                   |
 | -------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------- |
-| Agentera                   | The agent. One persistent identity that thinks through every step and keeps working when you walk away — the colleague, not the team. The CLI is the brain and spine: it owns routing, memory, capability prose, and the worker-spec contract. Active host coding agents (OpenCode, Codex, Cursor, Copilot) are harnesses that provide the model and tool runtime; Agentera provides the identity, memory, discipline, and routing. Identity is per-project: same personality, same capabilities, different memory, no bleed. Headline: one agent, one CLI, many capabilities. | `README.md`, `.agentera/vision.yaml`, `.agentera/decisions.yaml` (D74), `skills/agentera/SKILL.md` |
-| Host runtime (harness)      | An active host coding agent — OpenCode, Codex, Cursor, or Copilot — that provides the model, the tool runtime, and the spawn capacity. Agentera is the agent; the host is the harness. The host learns one contract: the JSON the CLI returns from `agentera prime`. Editor runtime installs are transport (the `skills/` directory, `plugin.json`), not an extension surface. | `.agentera/vision.yaml`, `.agentera/decisions.yaml` (D74), `skills/agentera/SKILL.md` |
-| Product surface            | Where the same fixed Agentera workflow ships: `@agentera/cli` (the agent runtime and narrow waist — owns routing, memory, capability prose, and the worker-spec contract), `@agentera/mobile` (primary app, a client of the CLI), `@agentera/web` (site and docs), and active host runtimes (OpenCode, Codex, Cursor, Copilot). Host runtimes are harnesses — clients of the CLI, not extension hosts or peers. | `README.md`, `AGENTS.md`, `packages/*/README.md`, `.agentera/decisions.yaml` (D74) |
+| Agentera                   | The agent. One persistent identity that thinks through every step and keeps working when you walk away — the colleague, not the team. The CLI is the brain and spine: it owns routing, memory, capability prose, and the worker-spec contract. Coding hosts consume the same shared skill and CLI contract; Agentera does not ship host-native package surfaces. Identity is per-project: same personality, same capabilities, different memory, no bleed. Headline: one agent, one CLI, many capabilities. | `README.md`, `.agentera/vision.yaml`, `.agentera/decisions.yaml` (D74), `skills/agentera/SKILL.md` |
+| Host runtime (harness)      | A coding agent host that provides the model, tools, and worker execution. Agentera is the agent; the host is the harness. Hosts consume the shared skill at `~/.agents/skills/agentera` and the JSON returned by `agentera prime`; Agentera does not install or repair host-native plugins, hooks, agents, or packages. | `.agentera/vision.yaml`, `.agentera/decisions.yaml` (D74), `skills/agentera/SKILL.md` |
+| Product surface            | Where the fixed Agentera workflow ships: `@agentera/cli` (the agent runtime and narrow waist), `@agentera/mobile` (a client of the CLI), and `@agentera/web` (site and docs). Coding hosts consume the shared skill and CLI rather than receiving native Agentera package surfaces. | `README.md`, `AGENTS.md`, `packages/*/README.md`, `.agentera/decisions.yaml` (D74) |
 | @agentera/mobile           | The flagship mobile/web app package at `packages/mobile`. SvelteKit, Cursor SDK, Cloudflare Worker.                                                                                                                                                                                 | `packages/mobile/README.md`, `packages/mobile/DESIGN.md`                         |
-| Agentera skill             | Internal transport, not a product word. The bundle at `skills/agentera/` packages Agentera's routing entry and capability prose for host editor runtimes — the `skills/` directory and `plugin.json` manifests host editors require. It is the same fixed workflow, not user extensibility. Per Decision 74, "skill" is retired from product vocabulary and survives only as internal transport; the product word is capabilities. | `skills/agentera/SKILL.md`, `README.md` Internals, `.agentera/decisions.yaml` (D74) |
+| Agentera skill             | Shared transport, not a product extension point. The source at `skills/agentera/` is installed once at `~/.agents/skills/agentera` and packages Agentera's routing entry and capability prose without host-native manifests. It is the same fixed workflow, not user extensibility. Per Decision 74, the product word is capabilities. | `skills/agentera/SKILL.md`, `README.md` Internals, `.agentera/decisions.yaml` (D74) |
 | Capability                 | A routed behavioral unit of the single Agentera agent, driven through the CLI. Eleven top-level capabilities (status, vision, discuss, research, plan, build, optimize, audit, document, design, orchestrate) plus profile, which surfaces through existing capabilities but appears as a peer row in user-facing docs. Each capability has a prose module at `packages/cli/src/capabilities/<name>/instructions.ts` plus `triggers.yaml`, `artifacts.yaml`, `validation.yaml`, and `exit.yaml`. English names are canonical per Decision 70; the Swedish `-era` names are legacy v2 stable only (see Legacy Swedish capability names). | `AGENTS.md`, `skills/agentera/capabilities/*`, `packages/cli/src/capabilities/*`, `.agentera/decisions.yaml` (D70, D74) |
 | Capability canonical name (v3) | The English name binding for v3+ capability invocation, per Decision 70, under the single-agent identity crystallized in Decision 74. Promoted from the Decision 43 alias set. Mobile UX, web docs, host runtimes, and the CLI all use the same English name; the v2 stable distribution uses the legacy Swedish `-era` IDs (see `Legacy Swedish capability names`). | `.agentera/decisions.yaml` (D43, D70, D74), `references/cli/vocabulary-index.yaml` (protected_surfaces) |
 | Legacy Swedish capability names (v2 stable) | The historical Swedish `-era` IDs (e.g. `hej`, `resonera`) used by the v2 stable distribution (`npx -y agentera@latest`) and preserved as historical references in archived plans, decisions, and changelogs. Out of scope for v3 surface per Decisions 70 and 74; the single-agent identity uses English capability names exclusively. Coexistence probe surfaces per-distribution naming divergence. | `.agentera/decisions.yaml` (D70, D74), `references/cli/vocabulary-index.yaml` (protected_surfaces) |
@@ -142,14 +142,14 @@ authoritative recovery command.
 | Suggest, don't force         | Status recommends the next capability but waits for user confirmation.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        |
 | Capability handoff label     | A recommendation from one capability to another. Use glyph plus canonical name, such as `⧉ build` or `≡ plan`, not standalone slash-capability names. SG priority codes are internal protocol references and are not user-facing handoff labels.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
 | Explicit route documentation | User-facing examples that teach the actual entry route. Use `/agentera <alias>` such as `/agentera build`; do not present aliases as CLI commands.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                         |
-| Runtime question tool        | Host-native bounded-choice prompt. Concrete tool availability belongs to the active host runtime, not this vocabulary.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                      |
-| Question-tool gating         | Use a native question tool only for at least two meaningful non-terminal next actions or consequential Proceed/Cancel; `Done` and custom/free-form answers do not count as alternatives. Initial Agentera/status briefs stay free-form unless bounded choices were requested or the suggested next step is a state-changing Proceed/Cancel handoff. A single non-mutating suggested handoff may use a free-form prompt, but a single state-changing handoff uses native Proceed/Cancel confirmation. State-changing means the proposed next step may write artifacts, edit code, run optimization or orchestration cycles, apply migrations, refresh app/runtime state, or otherwise mutate project/runtime state. Apply the behavior rule first, with examples such as ⧉ build, ≡ plan when creating or updating plans, ▤ document when writing docs, ⎘ optimize when running or applying optimization cycles, and ⎈ orchestrate when dispatching cycles. This dispatcher rule governs status and capability handoff prompts; invoked capability prose can impose stricter question-tool requirements. |
+| Bounded-choice prompt        | A question surface supplied by the coding host. Agentera may use one when available but does not install, configure, or advertise a host-specific prompt integration.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          |
+| Question gating              | Use bounded choices only for at least two meaningful non-terminal next actions or consequential Proceed/Cancel; `Done` and custom/free-form answers do not count as alternatives. Initial Agentera/status briefs stay free-form unless bounded choices were requested or the suggested next step is a state-changing Proceed/Cancel handoff. A single non-mutating suggested handoff may use a free-form prompt, but a single state-changing handoff uses Proceed/Cancel confirmation. State-changing means the proposed next step may write artifacts, edit code, run optimization or orchestration cycles, apply migrations, or otherwise mutate project state. Apply the behavior rule first; invoked capability prose can impose stricter requirements. |
 | Handoff confirmation         | Clear free-form acceptance of the named single suggestion confirms invocation. Selecting a downstream capability option in a bounded prompt also confirms invocation; selecting `Done` stops without routing. Ambiguous replies get one clarifying question.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               |
 | Route                        | Direct user invocation by canonical capability name, primary alias, or slash route. A route is already consent to invoke the capability and does not need an extra handoff confirmation.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                   |
 | Suggest                      | Recommend a downstream capability and wait for confirmation.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               |
 | Delegate                     | Orchestrate assigns approved plan work to a worker capability during an explicitly orchestrated flow.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       |
-| Spawn                        | Build or Optimize launches an isolated runtime worker through the host subagent mechanism.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                             |
-| Subagent mechanism           | Runtime support for worker execution through OpenCode, Codex CLI, Cursor IDE, Copilot CLI, or another active host-native worker surface.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
+| Spawn                        | Build or Optimize asks the coding host to launch an isolated worker when the host provides that facility.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                            |
+| Worker mechanism             | A host-provided worker facility. Agentera ships no host-specific worker descriptor, adapter, plugin, hook, or package.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                   |
 | Legacy bridge                | Temporary v1 entry points, especially `/hej`, that guide users to `/agentera` and the v2 upgrade path.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                     |
 
 CLI-visible `agentera prime` labels are source labels. Preserve them in CLI tests
@@ -368,18 +368,14 @@ not add precision:
 | Missing normal directory     | Previewable. Agentera can show a no-write repair preview.                                                                                                                                                                                                      |
 | Missing chosen directory     | Needs a user decision when provided through `AGENTERA_HOME` or explicit `--install-root`.                                                                                                                                                                      |
 | Package refresh              | Package-manager or marketplace update. It does not prove Agentera's app files are current.                                                                                                                                                                     |
-| App repair                   | Normal repair flow that previews or applies Agentera app files plus managed runtime config, plugins, hooks, commands, and safe cleanup together. It must not edit shell startup files, and package-manager commands remain opt-in through `--update-packages`. |
-| `--only bundle`              | Compatibility selector for narrow app-file work. Do not present it as the normal repair recommendation when managed runtime surfaces may also need repair.                                                                                                     |
+| App repair                   | Normal flow that previews or applies Agentera app/project work. It does not install, update, authenticate, enable, trust, or repair host-native resources.                                                                                                     |
 | Preview                      | No-write mode. Required before upgrade or app repair writes; the underlying command flag is `--dry-run`.                                                                                                                                                       |
 | `--yes`                      | Explicit apply flag after preview and approval.                                                                                                                                                                                                                |
-| Final check                  | Setup validation after upgrade apply. Uses the same app-home probe as `agentera doctor` (`build_doctor_status`), not `setup_doctor.build_report`.                                                                                                              |
-| Package-update opt-in        | External package manager changes require `--update-packages`.                                                                                                                                                                                                  |
-| Runtime adapter              | Runtime-specific Agentera adapter support for skill loading, hooks, artifact validation, lifecycle metadata, and diagnostics.                                                                                                                                  |
+| Final check                  | Validation after upgrade apply using the same app/project evidence reported by `agentera doctor`.                                                                                                                                                              |
 | Host support                 | What a runtime can theoretically do. Distinguish it from shipped Agentera behavior.                                                                                                                                                                            |
-| Hook lifecycle               | Runtime callbacks such as `SessionStart`, `Stop`, `PreToolUse`, and `PostToolUse`.                                                                                                                                                                             |
-| Setup doctor                 | Diagnostic command surface for install/runtime health.                                                                                                                                                                                                         |
+| Retired native integration vocabulary | Runtime selectors, adapter repair, plugin-shipped hooks, setup-doctor runtime health, and package-update flags describe released history or migration evidence only. They are not current Agentera features.                                                                                                            |
 
-Canonical active runtime names are OpenCode, Codex, Cursor, and GitHub Copilot. Cursor Agent CLI and Cursor IDE are surfaces of the single Cursor identity; `cursor-agent` is an inactive Cursor alias and source-product name. Claude Code is a retired migration and consent-gated historical-import source, not an active runtime.
+Claude Code is a retired migration and consent-gated historical-import source, not an active runtime. Current coding hosts consume the shared skill and CLI without entering an Agentera runtime inventory.
 
 ### App lifecycle status vocabulary
 
@@ -452,12 +448,11 @@ identifiers, package metadata, historical records, fixtures, path-like
 references, generic plain language, and ambiguous current prose.
 
 Use this prose as guidance only: current conceptual docs should say the object
-they mean, such as Agentera app files, suite package, plugin-shipped hooks,
-removed `bundle-status` command, Agentera routing entry point, skill entry file,
-historical v1 skill entry paths (post-3.0 removed from the repo tree). Preserve shipped identifiers and
-literal paths such as `.agentera-bundle.json`, `bundle.status`,
-`activeBundleRoot`, `--only bundle`, and `skills/agentera/SKILL.md` unless an
-explicit compatibility migration is in scope.
+they mean, such as Agentera app files, suite package, Agentera routing entry
+point, or skill entry file. Plugin-shipped hooks and `--only bundle` are retired
+v2 vocabulary and may appear only in released history, fixtures, or explicit
+migration evidence. Literal current paths such as `skills/agentera/SKILL.md`
+remain valid.
 
 Do not replace this with a parallel Markdown table of allowed and forbidden
 terms. Update the YAML authority first, then keep this section as the short
@@ -530,7 +525,7 @@ schema concept, runtime capability, install state, or user action.
 | Ambiguous term  | Why bare usage is risky                                                                                | Required wording                                                                                                                                                                     |
 | --------------- | ------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
 | Skill           | Confuses v1 standalone skills, the host editor skill bundle, and internal workflows.                  | Per Decision 74, `skill` is internal transport, not product vocabulary. Use `capability` for Agentera workflows; reserve `skill` for the `skills/` directory and host editor manifests; `v1 skill` for history.       |
-| Contract        | Could mean schema structure, artifact shape, protocol primitive, adapter behavior, or product promise. | Use `schema contract`, `artifact schema`, `protocol primitives`, `runtime adapter contract`, or `product promise`.                                                                   |
+| Contract        | Could mean schema structure, artifact shape, protocol primitive, package behavior, or product promise. | Use `schema contract`, `artifact schema`, `protocol primitives`, `package contract`, or `product promise`.                                                                           |
 | Status          | Different surfaces use different state machines and output labels.                                     | Use `exit status`, `task status`, `installed-app status`, `install status`, `docs status`, or `health status`.                                                                       |
 | Freshness       | Sounds like a branded synonym for several normal states: current, stale, synced, or out of date.       | Use object-specific state wording such as `artifact is current`, `Agentera app files need repair`, `docs are current`, or `plan-level current-state check`.                          |
 | Checkpoint      | In software, can mean commit, savepoint, restore point, model checkpoint, or milestone.                | Use `final state sync`, `plan-level current-state check`, `checkpoint commit`, or `pre-dispatch checkpoint commit`.                                                                  |

package/bundle/references/meta/documentation-inventory.md

@@ -10,7 +10,7 @@ guides — not under `docs/`.
 
 | Path | Topic |
 | ---- | ----- |
-| `docs/packaging/v3-packaging.md` | v3 CLI npm/Bun distribution contract |
+| `docs/packaging/v3-packaging.md` | v3 CLI npm distribution and verification-lane contract |
 | `docs/consolidation/monorepo-plan.md` | Product pivot, `@agentera/*` package layout, migration checklist (D67) |
 
 ## Authority stack

package/bundle/skills/agentera/schemas/artifacts/docs.yaml

@@ -73,10 +73,10 @@ BUDGET:
   1:
     id: DB1
     scope: full_file
-    max_words: 2000
+    max_words: null
     description: >-
-      Maximum words for the full file. If a write would exceed this
-      budget, compact older audit log entries.
+      No full-file word budget. Historical audit_log entries are immutable;
+      keep the active inventory and coverage sections concise.
 
 HEADER:
   1:
@@ -396,11 +396,3 @@ VALIDATION:
       Index status values should be from the allowed set.
     checks:
       - "Each status is one of: current, stale, missing, intent, generated"
-  4:
-    id: DVL4
-    rule: word_budget
-    severity: advisory
-    description: >-
-      Advisory check. Full file should be <=2000 words.
-    checks:
-      - "Total file word count <= 2000"