feat: publish portable mission runtime and sync boundary (#15)

Publishes the portable mission runtime, public synchronization boundary checks, package provenance and isolated installation/package smoke coverage. All required CI and release checks passed before merge.

Author: jcarlosrodicio

README.md

@@ -407,6 +407,12 @@ An explicit `--target` wins over a non-empty `OPENCODE_CONFIG_DIR`, which wins o
 
 The installer inventories the payload, computes a complete plan, and tracks ownership in `TARGET/.oak/manifest.json`. A dry run performs no writes. A real operation recomputes the plan under an exclusive lock and commits through a durable journal with one rollback generation.
 
+Release packages also record a safe `release_provenance` block in that manifest:
+the source projection commit, public commit, canonical kit version, packaged
+payload digest, and a passed-checks marker. Existing manifests without that
+optional block remain readable; the release package smoke requires it for new
+release evidence and never stores paths, provider settings, or credentials.
+
 The root `package.json` is the canonical kit-version source. Display the same
 identity through the manager or any lifecycle wrapper:
 
@@ -724,7 +730,10 @@ This runs the fast contract checker and every bundled `node:test` suite. For a
 fast structural check while editing documentation or contracts, use
 `npm run check:quick`. Before a release, use `npm run check:release`; it performs
 a frozen dependency install and also runs typechecking, dependency integrity,
-audit/signature checks, installation smoke, and the exact package smoke.
+audit/signature checks, installation smoke, and the exact package smoke. The
+package smoke extracts the tarball, recomputes the packaged OpenCode payload
+digest, uses an isolated `OPENCODE_CONFIG_DIR`, preserves a user configuration,
+and exercises `doctor` plus rollback before reporting release provenance.
 Publication remains separately authorized; follow the reviewed artifact and
 checksum procedure in [the supply-chain policy](docs/supply-chain.md).
 

docs/performance/native-tools-benchmark.md

@@ -0,0 +1,33 @@
+# Native tool-path benchmark
+
+This benchmark measures the current dependency-free Node path before deciding
+whether a native Rust helper is justified. It is evidence collection, not a
+Rust implementation or a release blocker.
+
+Run it with at least two iterations per operation:
+
+```bash
+node scripts/benchmark-tool-paths.mjs --iterations 5 > /tmp/oak-native-benchmark.json
+```
+
+It exercises two synthetic fixtures and, by default, this repository. Each
+fixture measures file listing, text search, Node parse checking, capability
+snapshot generation, and installation through the same manager used by the
+package. The report contains p50/p95 elapsed time, p50 CPU, and p95 RSS for
+each path. RSS is the benchmark runner's process value; child-command memory is
+not isolated, so the result must not be presented as a process-level peak for
+`node --check`.
+
+## Adoption threshold
+
+Keep the Node implementation unless a Rust replacement demonstrates, on both
+fixtures and two consecutive runs:
+
+- at least 2x lower p95 latency for one material hot path;
+- no regression above 10% in p95 RSS or installation time;
+- equivalent portable behavior and failure semantics;
+- a reviewable multi-platform distribution and checksum plan.
+
+The current repository has no Rust candidate or cross-platform artifact to
+compare. Until that evidence exists, adding Rust would increase distribution
+and supply-chain surface without a measured product benefit.

docs/sync-boundary.md

@@ -0,0 +1,27 @@
+# Public synchronization boundary
+
+This repository publishes a reviewed public projection of a fuller private
+harness. The public payload contains portable agents, commands, skills,
+contracts, tests, and read-only runtime helpers under `opencode/`, plus the
+installation and release tooling required to distribute them.
+
+The projection is deliberately not a private checkout mirror. Provider
+discovery, local model assignment, credentials, authentication files, MCP
+servers, private endpoints, sessions, transcripts, logs, and raw runtime
+evidence stay outside this repository. Public files must use portable relative
+paths and example service URLs only.
+
+<!-- projection-source-commit: 1ca607adcbae12cd16de1772662d942852b32502 -->
+
+The marker above identifies the reviewed source projection for release
+provenance. It is a commit identity only; it does not disclose the source
+checkout, its providers, or its runtime evidence.
+
+The source-side manifest classifies changes as `copy`, `translate`,
+`transform`, `exclude`, or `generated`. The private comparator verifies that
+classification before a change reaches this repository. This repository's
+`check-public-boundary.mjs` then verifies the required projection surfaces and
+reuses the same private-marker scan as `check.sh`.
+
+The boundary check is read-only. It does not install dependencies, publish a
+package, create a release, or synchronize a target checkout.

opencode/commands/loop-status.md

@@ -0,0 +1,24 @@
+---
+description: "Consulta el estado durable de una misión sin adquirir locks ni escribir archivos."
+agent: lead
+---
+
+# Estado de misión
+
+Usa esta consulta para inspeccionar un loop existente sin reanudarlo ni
+modificarlo.
+
+```bash
+node scripts/mission-status.mjs --root /ruta/absoluta/al/repo --slug <slug>
+node scripts/mission-status.mjs --root /ruta/absoluta/al/repo --slug <slug> --json
+```
+
+La salida proyecta el snapshot canónico y el último evento validado:
+`status`, `iteration/planned`, último paso, causa de bloqueo, sesión, timestamp
+y siguiente acción segura. `--root` debe ser una ruta absoluta y `--slug` un
+identificador kebab-case válido.
+
+Si falta el estado, el snapshot está desfasado o el historial está corrupto,
+detente y reporta el código durable (`state_missing`, `snapshot_stale` o
+`history_corrupt`). No ejecutes `resume`, `repair` ni `release` como sustituto
+automático.

opencode/commands/loop.md

@@ -16,6 +16,8 @@ automation.
 ```text
 approval_gate: explicit_before_writes
 max_iterations_per_invocation: 3
+planned_iteration_budget: task_specific_1_to_6
+hard_safety_ceiling: 6
 completion_authority: reviewer_only
 canonical_review_policy: code-review-and-quality/references/review-policy.md
 final_review_authority: reviewer
@@ -26,6 +28,12 @@ human_view_path: .opencode/loops/<slug>.md
 worktree_mode: explicit_opt_in
 ```
 
+The three-iteration value limits one invocation. The persisted
+`planned_iteration_budget` (`planned_iterations` in canonical state) is the total budget for the mission and may not
+exceed the hard ceiling of six; `current_iteration` may never exceed that
+budget. Once the total budget is exhausted, the loop must stop as `blocked` or
+`completed` rather than injecting another continuation.
+
 The durable `handoff_packet` (`.opencode/handoffs/<slug>.md` with
 `approval_status`) persists human approval to survive session restarts. `/loop`
 keeps `.opencode/loops/<slug>.json` as canonical state,

opencode/docs/ai/harness/agents.md

@@ -83,6 +83,14 @@ generate or modify frontmatter.
 - `evaluator`, `debugger`, and `evolver` are optional sidecars.
 - `evolver` works only on the OpenCode harness.
 
+## Mission runtime projection
+
+`mission-status` is a read-only view over the durable loop snapshot and
+append-only history. It validates canonical state before displaying progress
+and never becomes a second authority. The runtime observer is ephemeral; it
+may report activity, but durable transitions still require the loop state
+runtime and its locks.
+
 ## Bounded local autonomy
 
 `/autonomous` uses one explicit user invocation for a local objective, not a

opencode/docs/ai/harness/checks.md

@@ -64,6 +64,12 @@ Slice 2.4 adds durable structured state for `/loop`. Run its focused tests with:
 node --test scripts/loop-state.test.mjs
 ```
 
+The mission runtime projection is read-only: `mission-status.mjs` validates
+the canonical snapshot and history, while `mission-runtime` observes ephemeral
+events without writing an alternative state. The Open Design boundary requires
+`OPEN_DESIGN_URL` to be a base URL and rejects project or file URLs; the checker
+also rejects private endpoints and UUID-specific crypto dependencies.
+
 The dependency-free runtime maintains a JSON snapshot, append-only JSONL
 history, and exclusive lock. Tests cover crash recovery, same-session
 contention, symlink boundaries, journal continuity, migration, and explicit

opencode/docs/ai/harness/commands.md

@@ -174,6 +174,16 @@ Criteria:
 - No auto-merge, scheduling, write-enabled MCP connectors, parallel execution,
   or implicit worktree creation.
 
+## `/loop-status`
+
+Contract: `lead` read-only projection.
+
+`/loop-status --slug <slug>` inspects the canonical snapshot and append-only
+history without acquiring a write lock or changing files. Use
+`node scripts/mission-status.mjs --root /absolute/repository --slug <slug>`
+when an explicit repository root is needed. If state is missing, stale, or
+corrupt, report the durable error and do not repair or resume automatically.
+
 ## `/autonomous`
 
 Contract: explicit invocation -> `developer -> reviewer -> developer (state sync)`.

opencode/docs/ai/harness/orchestration-contracts.json

@@ -210,6 +210,12 @@
       "subtask": false,
       "workflow": "loop"
     },
+    {
+      "id": "loop-status",
+      "root_agent": "lead",
+      "subtask": false,
+      "workflow": "loop"
+    },
     {
       "id": "mvp-spec",
       "root_agent": "scoper",
@@ -643,7 +649,8 @@
     {
       "id": "loop",
       "entrypoints": [
-        "loop"
+        "loop",
+        "loop-status"
       ],
       "stages": [
         {
@@ -908,6 +915,7 @@
         "implement",
         "init",
         "loop",
+        "loop-status",
         "mvp-spec",
         "plan",
         "research",

opencode/docs/ai/specs/mission-runtime.md

@@ -0,0 +1,44 @@
+---
+title: "Proyección runtime de misión"
+status: proposed
+date: 2026-08-21
+feature: mission-runtime
+---
+
+# Runtime de misión
+
+## Autoridad
+
+`scripts/loop-state.mjs` es la única autoridad de ejecución. El snapshot
+`.opencode/loops/<slug>.json` y su historial append-only
+`.opencode/loops/<slug>.history.jsonl` se leen y validan; la proyección nunca
+los modifica ni mantiene un segundo estado durable.
+
+## Proyección read-only
+
+`mission-status.mjs` expone `slug`, `status`, `current_iteration`,
+`planned_iterations`, `last_completed_step`, `blocking_cause`, `session_id`,
+`updated_at` y `next_action`. `session_id` usa el lease activo cuando existe y
+el último session id canónico en caso contrario. `updated_at` procede del último
+evento validado del historial.
+
+Los estados canónicos son `approved`, `running`, `paused`, `blocked` y
+`completed`. La proyección no inventa transiciones: si el snapshot está
+desfasado o el historial está corrupto, devuelve el error durable de
+`loop-state`.
+
+## Observación de eventos
+
+El observador opcional usa únicamente eventos que el SDK local expone:
+`session.created`, `session.status`, `tool.execute.before`,
+`tool.execute.after`, `session.compacted`, `session.error` y `session.idle`.
+La actividad derivada (`running`, `idle` o `blocked`) es efímera y nunca
+sustituye al `status` canónico.
+
+Las sesiones con `parentID` distinto de la sesión raíz se ignoran. Los eventos
+duplicados se descartan por una clave efímera de sesión, tipo y payload; perder
+esa deduplicación al reiniciar no altera el snapshot ni el historial.
+
+La compaction conserva el session id. Un error de notificación no bloquea el
+loop ni se convierte en una transición durable. Las transiciones reales siguen
+requiriendo `loop-state.mjs` y sus locks, hashes, `action_id` y presupuesto.