pantheon-opencode 1.5.0-beta.6 → 1.5.0-beta.8

package/CHANGELOG.md

@@ -18,6 +18,83 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ## ✅ Closed Issues
 
+## [v1.5.0-beta.8] - 2026-09-13
+
+<!-- Add new changes here. Running `node scripts/versioning.mjs apply` will
+     move this section to a versioned entry and reset the template below. -->
+
+## 🐞 Fixed
+
+- **Board persistence is corruption- and concurrency-safe:** state writes now go
+  through a UNIQUE temp file (per-process, per-write counter + random suffix), are
+  `fsync`ed, then renamed over the destination with bounded ENOENT retries, and
+  read-modify-write cycles on the same state path are serialized. This closes the
+  window where two writers could interleave and corrupt `state.json` or race the
+  rename.
+- **Corrupt board state is quarantined and salvaged:** a malformed `state.json` is
+  moved aside to `<state>.corrupt-<ts>`, a clear error is logged, and every complete
+  record that can be parsed is salvaged and returned instead of throwing — so
+  delete/prune/recover callers cannot get stuck in a retry loop.
+- **P1-1 — no orphan `.signal.json` on reconcile:** reconcile now deletes any
+  still-pending wake signal instead of re-writing it. Reconcile is an
+  acknowledgment, not a new terminal transition, so the finalize→reconcile cycle
+  no longer leaves a stale `<alias>.signal.json` with `state:"reconciled"` behind.
+- **Unresolved `{sessionID}` template guard:** the TUI/runtime can hand the plugin
+  the unsubstituted literal `{sessionID}` (or `%7BsessionID%7D`) when no session is
+  focused; the plugin now refuses placeholder session ids at the delegation,
+  bootstrap-probe, and activity-sample boundaries instead of forwarding them to the
+  session API (which rejected every call).
+- **Delegation lookup scoped to the calling session:** `pantheon_delegation_read`
+  resolves exact task IDs first (must belong to the caller's session), then alias
+  matches within that session (exactly one required). Cross-session and ambiguous
+  aliases now return explicit errors instead of the generic "Unknown delegation".
+
+
+## [v1.5.0-beta.7] - 2026-09-13
+
+&lt;!-- Add new changes here. Running `node scripts/versioning.mjs apply` will
+     move this section to a versioned entry and reset the template below. --&gt;
+
+## 🆕 What's New
+
+- **Cost CLI fallback + `PANTHEON_NODE`:** `pantheon_cost` falls back to a
+  read-only `node scripts/cost.mjs` subprocess when `node:sqlite` is
+  unavailable in the host runtime (for example under Bun). The fallback
+  resolves a real Node.js binary — `PANTHEON_NODE` first, then `node` on
+  `PATH` — probes it for `node:sqlite` support (Node.js >= 22.5), and runs it
+  with an argument array, never a shell. Previously the tool returned
+  `status: UNSUPPORTED` with no fallback.
+- **Delegation engine contract (ADR-0011):** legacy `pantheon_delegate` remains
+  the default engine through v1.5.x; native delegation is opt-in and
+  experimental via `PANTHEON_DELEGATE_MODE=native` until v1.6. See
+  [ADR-0011](https://github.com/ils15/pantheon-opencode/pull/126) (Proposed).
+
+## 🐞 Fixed
+
+- **Legacy read-only bypass on retry (P0):** a retried delegation now
+  re-applies read-only enforcement (the same rule as the initial child) and
+  registers the retry session, closing the bypass where a retried child could
+  write despite `read_only: true` or being a read-only agent.
+- **Kill-switch scope (P0):** `PANTHEON_DELEGATION=off` now also disables the
+  legacy engine, not only the native manager — the legacy delegate/read/list
+  tools throw as documented.
+- **Delegation signal leak (P1):** the legacy finalize path now deletes the
+  auto-wake signal file, so terminal delegations no longer leave stale
+  `.signal.json` artifacts behind.
+- **Crash recovery is durable (P1):** `recoverRunningJobs` persists the
+  `error` state, marks the job `terminalUnreconciled`, and emits the terminal
+  signal plus notification for every reaped job instead of updating memory
+  only.
+- **Atomic concurrency (P1):** `registerLaunch` validates
+  `maxConcurrentPerAgent` before any mutation, so concurrent dispatches cannot
+  both claim the last slot. A refused launch tears down the orphan child
+  session (read-only unregister plus `session.delete`) and the retry path
+  re-validates the same limit.
+- **Legacy board hygiene (P1):** the legacy finalize path now runs
+  `pruneCompleted` and `enforceEntryCap`, matching the native manager and
+  preventing unbounded board growth.
+- **Resilient V2 events:** a native child refused by the per-agent limit is
+  logged and skipped instead of aborting the rest of the event handler.
 ## [v1.5.0-beta.6] - 2026-09-12
 
 &lt;!-- Add new changes here. Running `node scripts/versioning.mjs apply` will

package/README.md

@@ -76,22 +76,43 @@ availability and configuration of OpenCode and any optional services you choose
 to use. Check the [releases](https://github.com/ils15/pantheon-opencode/releases)
 and [changelog](CHANGELOG.md) for the latest published changes.
 
-## Native delegation mode
+## Delegation engines (legacy default, native opt-in)
 
-Set `PANTHEON_DELEGATE_MODE=native` to use OpenCode's native child-session
-delegation. Native mode is strict: it does not use foreground fallback, model
-failover, or retry; it performs a lazy capability probe on the first dispatch,
-and includes the resulting status in the delegation receipt.
+Pantheon exposes two delegation engines behind one tool surface:
+
+- **Legacy (default)** — `pantheon_delegate`, the original engine registered by
+  the V1 plugin. No configuration is required.
+- **Native (opt-in, experimental)** — OpenCode's native `task()` child-session
+  engine, selected with `PANTHEON_DELEGATE_MODE=native`. Native is strict: it
+  does not use foreground fallback, model failover, or retry; it performs a
+  lazy capability probe on the first dispatch and includes the resulting
+  status in the delegation receipt.
+
+Legacy stays the default for v1.5.x; native is not promoted to default until
+v1.6. See [ADR-0011](.pantheon/memory-bank/adr/0011-delegation-engine-contract.md)
+(Proposed) for the engine contract, rationale, and migration prerequisites.
+
+**Kill-switch:** set `PANTHEON_DELEGATION=off` (case-insensitive) to disable
+delegation entirely. The switch covers **both** engines — the legacy
+`pantheon_delegate`/read/list tools throw, and the native manager throws too.
 
 ## Cost tool backend
 
-`pantheon_cost` has one backend: read-only `node:sqlite` against the selected
-`opencode.db`. On Node versions without `node:sqlite` (Node < 22.5), the tool
-returns `status: UNSUPPORTED`; it has no CLI fallback. Database failures return
-a contract status such as `UNAVAILABLE` or
-`CORRUPT_DATA` and include diagnostic detail, including captured stderr when
+`pantheon_cost` prefers a read-only `node:sqlite` backend against the selected
+`opencode.db`. When `node:sqlite` is unavailable in the host runtime (for
+example under Bun), the tool falls back to spawning `node scripts/cost.mjs` as
+a read-only subprocess. The fallback resolves a real Node.js binary —
+`PANTHEON_NODE` first, then `node` on `PATH` — verifies it supports
+`node:sqlite` (Node.js >= 22.5), and runs the script with an argument array,
+never a shell. If neither source resolves, the tool returns `status:
+UNSUPPORTED`. Database failures return a contract status such as `UNAVAILABLE`
+or `CORRUPT_DATA` and include diagnostic detail, including captured stderr when
 available.
 
+| Variable | Default | Purpose |
+|----------|---------|---------|
+| `PANTHEON_NODE` | `node` from `PATH` | Path to a Node.js >= 22.5 binary used by the CLI fallback; an explicit path must exist and be executable, otherwise the tool fails fast instead of ignoring the override |
+
 
 ## Code-mode execution (explicit opt-in)
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pantheon-opencode",
-  "version": "1.5.0-beta.6",
+  "version": "1.5.0-beta.8",
   "description": "Multi-agent orchestration for OpenCode — 14 specialized agents, TDD, Themis review gates.",
   "type": "module",
   "publishConfig": {
@@ -41,8 +41,9 @@
     "test:security": "npm run secret-scan",
     "test:node": "node --test tests/*.mjs tests/pantheon/*.mjs",
     "test:ts": "for f in tests/pantheon/*.test.ts; do if [ \"$f\" = tests/pantheon/plugin-v2-contract.test.ts ]; then NODE_OPTIONS=--conditions=import tsx \"$f\"; else tsx \"$f\"; fi || exit 1; done",
-    "test": "python3 -m pytest tests/ -q 2>&1; status=$?; if [ \"$status\" -eq 5 ]; then echo \"No tests found\"; fi; exit \"$status\"",
-    "test:ci": "python3 -m pytest tests/ -q --tb=short 2>&1"
+    "test:all": "npm run test:ci && npm run test:node && npm run test:ts",
+    "test": "npm run test:all",
+    "test:ci": "python3 -m pytest -q --tb=short 2>&1"
   },
   "bin": {
     "pantheon-opencode": "bin/pantheon-init.mjs"

package/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "pantheon",
   "publisher": "ils15",
-  "version": "1.5.0-beta.6",
+  "version": "1.5.0-beta.8",
   "displayName": "Pantheon",
   "description": "Conductor-Delegate orchestration platform with 14 specialized agents, 21 skills, and OpenCode support.",
   "agents": "./src/agents",

package/pyproject.toml

@@ -4,13 +4,21 @@ build-backend = "setuptools.backends._legacy:_Backend"
 
 [project]
 name = "pantheon"
-version = "1.5.0-beta.6"
+version = "1.5.0-beta.8"
 description = "Multi-agent orchestration framework"
 requires-python = ">=3.11"
 dependencies = []
 
 [tool.pytest.ini_options]
-testpaths = ["tests"]
+# `tests/` holds the main suite; the two entries below are orphaned test
+# modules that live next to the code they cover and are collected here so
+# pytest (and therefore `npm test`) exercises them instead of letting them
+# rot as dead code.
+testpaths = [
+    "tests",
+    "scripts/vector_memory/test_memory.py",
+    "benchmarks/beta2/test_beta2.py",
+]
 asyncio_mode = "auto"
 addopts = "-v --tb=short"
 

package/scripts/mcp_persistence_server.py

@@ -777,13 +777,32 @@ def _validate_heartbeat(heartbeat: object) -> None:
         raise ValueError("heartbeat.turn_count must be a non-negative integer")
 
 
+def _validate_version_metadata(version: object) -> None:
+    """Validate optional checkpoint ``version`` metadata leniently.
+
+    ``version`` is opaque application metadata (rehydrate does not consume it),
+    not a persistence requirement. It may be a positive integer, a JSON float
+    such as ``1.0``, or a version label like ``"1.5.0-beta.2"``. Absent/null
+    versions and any non-empty string remain valid, preserving existing integer
+    checkpoints while allowing forward-compatible labels. Only bools, empty
+    strings, non-finite/zero/negative numbers, and structured values are
+    rejected.
+    """
+    if version is None:
+        return
+    if isinstance(version, bool):
+        raise ValueError("version must be a positive integer or version label")
+    if isinstance(version, (int, float)):
+        if not 1 <= version < float("inf"):
+            raise ValueError("version must be a positive integer or version label")
+        return
+    if not isinstance(version, str) or not version.strip():
+        raise ValueError("version must be a positive integer or version label")
+
+
 def _validate_checkpoint_shape(checkpoint: dict) -> None:
     """Validate bounded structured checkpoint fields without sanitizing them."""
-    version = checkpoint.get("version")
-    if version is not None and (
-        isinstance(version, bool) or not isinstance(version, int) or version < 1
-    ):
-        raise ValueError("version must be a positive integer")
+    _validate_version_metadata(checkpoint.get("version"))
     validators = (
         ("goal", _validate_goal),
         ("phase", _validate_phase),

package/scripts/versioning.mjs

@@ -21,8 +21,8 @@
  */
 
 import { execFileSync } from 'node:child_process'
-import { readFileSync } from 'node:fs'
-import { dirname, join } from 'node:path'
+import { existsSync, readFileSync } from 'node:fs'
+import { basename, dirname, join } from 'node:path'
 import { fileURLToPath } from 'node:url'
 import {
   commitPreparedUpdates,
@@ -37,6 +37,8 @@ const __dirname = dirname(fileURLToPath(import.meta.url))
 const ROOT = join(__dirname, '..')
 
 const CHANGELOG_PATH = join(ROOT, 'CHANGELOG.md')
+const ZENODO_PATH = join(ROOT, '.zenodo.json')
+const CITATION_PATH = join(ROOT, 'CITATION.cff')
 
 // ---------------------------------------------------------------------------
 // Git helpers
@@ -111,6 +113,90 @@ function prepareManifests(newVersion) {
   return { inventory, updates: prepareInventoryVersion(inventory, newVersion) }
 }
 
+/**
+ * Serialize `.zenodo.json` the way Biome 2.x formats it: objects stay expanded
+ * (matching `JSON.stringify(value, null, 2)`), while arrays containing only
+ * primitives are collapsed onto one line when they fit within the line width.
+ * A plain `JSON.stringify` emits the `keywords` array multiline, which Biome
+ * collapses — reintroducing a lint failure on the next release bump.
+ *
+ * @param {unknown} value parsed JSON document
+ * @param {number} [lineWidth] formatter line width (matches biome.json)
+ * @returns {string} Biome-compatible JSON with a trailing newline
+ */
+function formatZenodoJson(value, lineWidth = 100) {
+  const indent = (depth) => '  '.repeat(depth)
+  const render = (node, depth, column) => {
+    if (Array.isArray(node)) {
+      if (node.length === 0) return '[]'
+      const primitiveOnly = node.every((item) => item === null || typeof item !== 'object')
+      if (primitiveOnly) {
+        const inline = `[${node.map((item) => JSON.stringify(item)).join(', ')}]`
+        if (column + inline.length <= lineWidth) return inline
+      }
+      const body = node
+        .map((item) => `${indent(depth + 1)}${render(item, depth + 1, indent(depth + 1).length)}`)
+        .join(',\n')
+      return `[\n${body}\n${indent(depth)}]`
+    }
+    if (node !== null && typeof node === 'object') {
+      const entries = Object.entries(node)
+      if (entries.length === 0) return '{}'
+      const body = entries
+        .map(([key, item]) => {
+          const keyPrefix = `${indent(depth + 1)}${JSON.stringify(key)}: `
+          return `${keyPrefix}${render(item, depth + 1, keyPrefix.length)}`
+        })
+        .join(',\n')
+      return `{\n${body}\n${indent(depth)}}`
+    }
+    return JSON.stringify(node)
+  }
+  return `${render(value, 0, 0)}\n`
+}
+
+/**
+ * Keep the Zenodo deposition metadata (.zenodo.json) and the citation file
+ * (CITATION.cff) aligned with the version and date being released. Zenodo's
+ * release validator hard-fails when CITATION.cff version differs from the
+ * release tag, so these must be bumped together with the manifests.
+ */
+function prepareReleaseMetadata(newVersion, dateStr) {
+  if (!SEMVER_PATTERN.test(newVersion)) throw new Error(`invalid version: ${newVersion}`)
+  if (!/^\d{4}-\d{2}-\d{2}$/.test(dateStr)) throw new Error(`invalid date: ${dateStr}`)
+  const tagUrl = `https://github.com/ils15/pantheon-opencode/releases/tag/v${newVersion}`
+  const updates = []
+
+  if (existsSync(ZENODO_PATH)) {
+    const zenodo = JSON.parse(readFileSync(ZENODO_PATH, 'utf8'))
+    const zenodoPrevious = typeof zenodo.version === 'string' ? zenodo.version : null
+    if (zenodoPrevious && zenodoPrevious !== newVersion && typeof zenodo.description === 'string') {
+      zenodo.description = zenodo.description.replaceAll(zenodoPrevious, newVersion)
+    }
+    zenodo.version = newVersion
+    zenodo.publication_date = dateStr
+    zenodo.url = tagUrl
+    updates.push({ path: ZENODO_PATH, content: formatZenodoJson(zenodo) })
+  }
+
+  if (existsSync(CITATION_PATH)) {
+    const citationRaw = readFileSync(CITATION_PATH, 'utf8')
+    const citationMatch = /^version:\s*["']?([^"'\s]+)["']?\s*$/m.exec(citationRaw)
+    const citationPrevious = citationMatch ? citationMatch[1] : null
+    let citation = citationRaw
+    if (citationPrevious && citationPrevious !== newVersion) {
+      citation = citation.replaceAll(`Pantheon v${citationPrevious}`, `Pantheon v${newVersion}`)
+    }
+    citation = citation
+      .replace(/^version:.*$/m, `version: ${newVersion}`)
+      .replace(/^date-released:.*$/m, `date-released: ${dateStr}`)
+      .replace(/^url:.*$/m, `url: ${tagUrl}`)
+    updates.push({ path: CITATION_PATH, content: citation })
+  }
+
+  return updates
+}
+
 // ---------------------------------------------------------------------------
 // CHANGELOG updater
 //
@@ -243,8 +329,8 @@ function contentHasUnreleased() {
   return readFileSync(CHANGELOG_PATH, 'utf8').includes('## [Unreleased]')
 }
 
-function commitPreparedRelease(manifestPlan, changelogPlan) {
-  const updates = [...manifestPlan.updates]
+function commitPreparedRelease(manifestPlan, changelogPlan, metadataUpdates = []) {
+  const updates = [...manifestPlan.updates, ...metadataUpdates]
   if (changelogPlan.changed) {
     updates.push({ path: CHANGELOG_PATH, content: changelogPlan.content })
   }
@@ -252,6 +338,9 @@ function commitPreparedRelease(manifestPlan, changelogPlan) {
   for (const document of manifestPlan.inventory.documents) {
     console.log(`  ✓ ${document.entry.file} → ${manifestPlan.version}`)
   }
+  for (const { path } of metadataUpdates) {
+    console.log(`  ✓ ${basename(path)} → ${manifestPlan.version}`)
+  }
   if (changelogPlan.changed) {
     console.log(
       changelogPlan.notesGenerated
@@ -314,7 +403,11 @@ switch (command) {
       console.log(`Bumping ${current} → ${newVersion} (beta)`)
       const changelogPlan = prepareUnreleased(newVersion, date)
       const manifestPlan = prepareManifests(newVersion)
-      commitPreparedRelease({ ...manifestPlan, version: newVersion }, changelogPlan)
+      commitPreparedRelease(
+        { ...manifestPlan, version: newVersion },
+        changelogPlan,
+        prepareReleaseMetadata(newVersion, date),
+      )
       console.log(`\nDone. Commit the version inventory and CHANGELOG as v${newVersion}.`)
       console.log(`Tag v${newVersion} will be created by the release workflow after merge to main.`)
       break
@@ -347,7 +440,11 @@ switch (command) {
       const notesBody = generateNotes()
       const changelogPlan = prepareUnreleased(current, date, notesBody)
       const manifestPlan = prepareManifests(current)
-      commitPreparedRelease({ ...manifestPlan, version: current }, changelogPlan)
+      commitPreparedRelease(
+        { ...manifestPlan, version: current },
+        changelogPlan,
+        prepareReleaseMetadata(current, date),
+      )
       console.log(`Tag v${current} will be created by the release workflow after merge to main.`)
       break
     }
@@ -360,7 +457,11 @@ switch (command) {
     const notesBody = generateNotes()
     const changelogPlan = prepareUnreleased(newVersion, date, notesBody)
     const manifestPlan = prepareManifests(newVersion)
-    commitPreparedRelease({ ...manifestPlan, version: newVersion }, changelogPlan)
+    commitPreparedRelease(
+      { ...manifestPlan, version: newVersion },
+      changelogPlan,
+      prepareReleaseMetadata(newVersion, date),
+    )
     console.log(`\nDone. Commit the version inventory and CHANGELOG as v${newVersion}.`)
     console.log(`Tag v${newVersion} will be created by the release workflow after merge to main.`)
     break

package/src/agents/mnemosyne.md

@@ -34,7 +34,7 @@ permission:
   bash: deny
   read: allow
   grep: allow
-  edit: deny
+  edit: allow
   task:
     "*": deny
 ---

package/src/mcp/mcp_persistence_server.py

@@ -777,13 +777,32 @@ def _validate_heartbeat(heartbeat: object) -> None:
         raise ValueError("heartbeat.turn_count must be a non-negative integer")
 
 
+def _validate_version_metadata(version: object) -> None:
+    """Validate optional checkpoint ``version`` metadata leniently.
+
+    ``version`` is opaque application metadata (rehydrate does not consume it),
+    not a persistence requirement. It may be a positive integer, a JSON float
+    such as ``1.0``, or a version label like ``"1.5.0-beta.2"``. Absent/null
+    versions and any non-empty string remain valid, preserving existing integer
+    checkpoints while allowing forward-compatible labels. Only bools, empty
+    strings, non-finite/zero/negative numbers, and structured values are
+    rejected.
+    """
+    if version is None:
+        return
+    if isinstance(version, bool):
+        raise ValueError("version must be a positive integer or version label")
+    if isinstance(version, (int, float)):
+        if not 1 <= version < float("inf"):
+            raise ValueError("version must be a positive integer or version label")
+        return
+    if not isinstance(version, str) or not version.strip():
+        raise ValueError("version must be a positive integer or version label")
+
+
 def _validate_checkpoint_shape(checkpoint: dict) -> None:
     """Validate bounded structured checkpoint fields without sanitizing them."""
-    version = checkpoint.get("version")
-    if version is not None and (
-        isinstance(version, bool) or not isinstance(version, int) or version < 1
-    ):
-        raise ValueError("version must be a positive integer")
+    _validate_version_metadata(checkpoint.get("version"))
     validators = (
         ("goal", _validate_goal),
         ("phase", _validate_phase),

package/src/pantheon/background-job-board.ts

@@ -222,15 +222,28 @@ export class BackgroundJobBoard {
   async recoverRunningJobs(): Promise<void> {
     if (!this.persistence) return
     const records = await this.persistence.loadAllJobs()
+    const recoveredIDs: string[] = []
     for (const record of records) {
       if (record.state === 'running') {
         record.state = 'error'
         record.lastStatusError = 'Process restarted — job marked as errored'
         record.updatedAt = Date.now()
+        record.completedAt = Date.now()
         record.totalErrors++
+        record.terminalUnreconciled = true
+        recoveredIDs.push(record.taskID)
       }
       this.jobs.set(record.taskID, record)
     }
+    // Persist + signal + notify for each recovered job (write-ahead-log pattern)
+    for (const taskID of recoveredIDs) {
+      const job = this.jobs.get(taskID)
+      if (job) {
+        await this.persistRecord(job)
+        await this.writeSignal(job)
+        this.notifyTerminal(job.taskID)
+      }
+    }
   }
 
   // ─── Core Operations ────────────────────────────────────────────────
@@ -240,6 +253,16 @@ export class BackgroundJobBoard {
    * The job starts in the `running` state with a generated alias.
    */
   async registerLaunch(input: LaunchInput): Promise<BackgroundJobRecord> {
+    // Atomic concurrency check: validate BEFORE any mutation or await.
+    // JavaScript's single-threaded event loop guarantees this check + set
+    // cannot be interleaved with another registerLaunch call.
+    if (!this.canDispatch(input.agent)) {
+      throw new Error(
+        `Concurrency limit reached for agent "${input.agent}" ` +
+          `(${this.getRunningCount(input.agent)} running, max ${this.options.maxConcurrentPerAgent})`,
+      )
+    }
+
     const prefix = getAgentPrefix(input.agent)
     let sessionCounters = this.aliasCounters.get(input.parentSessionID)
     if (!sessionCounters) {
@@ -434,7 +457,12 @@ export class BackgroundJobBoard {
     job.terminalUnreconciled = false
 
     await this.persistRecord(job)
-    await this.writeSignal(job)
+    // P1-1: reconcile is an acknowledgment, NOT a new terminal transition
+    // (onTerminal does not re-fire). Re-writing a `.signal.json` here leaked
+    // orphan signals with `state:"reconciled"` after the finalize→reconcile
+    // cycle deleted them. Instead, remove any still-pending wake signal so
+    // reconcile leaves no `<alias>.signal.json` behind.
+    await this.deleteSignal(job.alias)
 
     return job
   }