polygram 0.22.0 → 0.23.1

package/migrations/015-agent-runtime-sessions.sql

@@ -0,0 +1,50 @@
+-- Provider/session compatibility namespaces.
+--
+-- The legacy sessions table remains unchanged for rollback. New binaries
+-- dual-write Claude rows during the rollback window. Codex never writes the
+-- legacy claude_session_id column.
+
+CREATE TABLE agent_runtime_sessions (
+  session_key TEXT NOT NULL,
+  namespace TEXT NOT NULL
+    CHECK(namespace IN ('claude:inline', 'claude:channels', 'codex:app-server')),
+  provider TEXT NOT NULL CHECK(provider IN ('claude', 'codex')),
+  provider_session_id TEXT NOT NULL,
+  app_server_session_id TEXT,
+  agent TEXT,
+  cwd TEXT,
+  model TEXT,
+  effort TEXT,
+  pm_backend TEXT,
+  created_ts INTEGER NOT NULL,
+  last_active_ts INTEGER NOT NULL,
+  PRIMARY KEY(session_key, namespace)
+);
+
+CREATE INDEX idx_agent_runtime_sessions_provider
+  ON agent_runtime_sessions(provider, namespace, last_active_ts DESC);
+
+-- Stored tmux rows are the historical inline backend. Only cli/channels rows
+-- carry the Channels reply-tool compatibility contract.
+INSERT INTO agent_runtime_sessions (
+  session_key, namespace, provider, provider_session_id,
+  app_server_session_id, agent, cwd, model, effort, pm_backend,
+  created_ts, last_active_ts
+)
+SELECT
+  session_key,
+  CASE
+    WHEN pm_backend IN ('cli', 'channels') THEN 'claude:channels'
+    ELSE 'claude:inline'
+  END,
+  'claude',
+  claude_session_id,
+  NULL,
+  agent,
+  cwd,
+  model,
+  effort,
+  pm_backend,
+  created_ts,
+  last_active_ts
+FROM sessions;

package/migrations/016-agent-turn-attempts.sql

@@ -0,0 +1,211 @@
+-- Durable Codex generation, request-delivery, effect, and recovery ledger.
+-- Payload-bearing command/tool data is intentionally absent.
+
+CREATE TABLE codex_runtime_identity (
+  singleton INTEGER PRIMARY KEY CHECK(singleton = 1),
+  stable_host_id TEXT NOT NULL,
+  last_boot_session_id TEXT NOT NULL,
+  established_ts INTEGER NOT NULL,
+  updated_ts INTEGER NOT NULL
+);
+
+CREATE TABLE codex_reboot_releases (
+  stable_host_id TEXT NOT NULL,
+  incident_boot_session_id TEXT NOT NULL,
+  released_boot_session_id TEXT NOT NULL,
+  released_ts INTEGER NOT NULL,
+  PRIMARY KEY(stable_host_id, incident_boot_session_id)
+);
+
+CREATE TABLE codex_generations (
+  generation_id TEXT PRIMARY KEY,
+  session_key TEXT NOT NULL,
+  thread_id TEXT,
+  app_server_session_id TEXT,
+  stable_host_id TEXT NOT NULL,
+  boot_session_id TEXT NOT NULL,
+  state TEXT NOT NULL
+    CHECK(state IN (
+      'created', 'active', 'healthy-stopped', 'containment-failed',
+      'durability-blocked', 'retired'
+    )),
+  containment_reason TEXT,
+  created_ts INTEGER NOT NULL,
+  updated_ts INTEGER NOT NULL,
+  settled_ts INTEGER
+);
+
+CREATE INDEX idx_codex_generations_state
+  ON codex_generations(state, updated_ts);
+CREATE INDEX idx_codex_generations_session
+  ON codex_generations(session_key, created_ts DESC);
+
+CREATE TABLE codex_daemon_lease (
+  singleton INTEGER PRIMARY KEY CHECK(singleton = 1),
+  generation_id TEXT REFERENCES codex_generations(generation_id)
+    ON DELETE SET NULL,
+  stable_host_id TEXT NOT NULL,
+  boot_session_id TEXT NOT NULL,
+  status TEXT NOT NULL CHECK(status IN ('clear', 'active', 'quarantined')),
+  quarantine_reason TEXT,
+  acquired_ts INTEGER,
+  updated_ts INTEGER NOT NULL,
+  released_ts INTEGER
+);
+
+CREATE TABLE codex_turn_attempts (
+  attempt_id TEXT PRIMARY KEY,
+  generation_id TEXT NOT NULL
+    REFERENCES codex_generations(generation_id) ON DELETE CASCADE,
+  session_key TEXT NOT NULL,
+  method TEXT NOT NULL,
+  thread_id TEXT,
+  turn_id TEXT,
+  telegram_source_message_id TEXT,
+  client_user_message_id TEXT,
+  request_id TEXT,
+  delivery_state TEXT NOT NULL
+    CHECK(delivery_state IN (
+      'prepared', 'write-attempted', 'response-observed'
+    )),
+  response_outcome TEXT CHECK(response_outcome IN ('result', 'error')),
+  recovery_state TEXT NOT NULL
+    CHECK(recovery_state IN (
+      'prepared', 'ambiguous', 'active', 'terminal-pending',
+      'clean-pending', 'empty-registry-pending', 'settled', 'cancelled'
+    )),
+  terminal_status TEXT
+    CHECK(terminal_status IN ('completed', 'interrupted', 'failed')),
+  created_ts INTEGER NOT NULL,
+  updated_ts INTEGER NOT NULL,
+  settled_ts INTEGER,
+  ambiguous_ts INTEGER
+);
+
+CREATE INDEX idx_codex_attempts_recovery
+  ON codex_turn_attempts(recovery_state, updated_ts);
+CREATE INDEX idx_codex_attempts_generation
+  ON codex_turn_attempts(generation_id, created_ts);
+
+CREATE TABLE codex_attempt_checkpoints (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  generation_id TEXT NOT NULL
+    REFERENCES codex_generations(generation_id) ON DELETE CASCADE,
+  attempt_id TEXT
+    REFERENCES codex_turn_attempts(attempt_id) ON DELETE CASCADE,
+  kind TEXT NOT NULL,
+  thread_id TEXT,
+  turn_id TEXT,
+  request_id TEXT,
+  item_id TEXT,
+  detail_json TEXT,
+  ts INTEGER NOT NULL
+);
+
+CREATE INDEX idx_codex_checkpoints_attempt
+  ON codex_attempt_checkpoints(attempt_id, id);
+CREATE INDEX idx_codex_checkpoints_generation
+  ON codex_attempt_checkpoints(generation_id, id);
+
+-- Provider choice is recorded before provider dispatch so replay never infers
+-- a backend from mutable current configuration. Historical rows without a
+-- selection remain unknown.
+CREATE TABLE inbound_runtime_selections (
+  bot_name TEXT NOT NULL,
+  telegram_chat_id TEXT NOT NULL,
+  telegram_message_id TEXT NOT NULL,
+  session_key TEXT NOT NULL,
+  provider TEXT NOT NULL CHECK(provider IN ('claude', 'codex')),
+  selected_ts INTEGER NOT NULL,
+  PRIMARY KEY(bot_name, telegram_chat_id, telegram_message_id)
+);
+
+CREATE INDEX idx_inbound_runtime_session
+  ON inbound_runtime_selections(session_key, provider, selected_ts);
+
+CREATE TABLE codex_linked_inputs (
+  linked_input_id TEXT PRIMARY KEY,
+  generation_id TEXT NOT NULL
+    REFERENCES codex_generations(generation_id) ON DELETE CASCADE,
+  attempt_id TEXT NOT NULL
+    REFERENCES codex_turn_attempts(attempt_id) ON DELETE CASCADE,
+  target_attempt_id TEXT NOT NULL
+    REFERENCES codex_turn_attempts(attempt_id) ON DELETE CASCADE,
+  telegram_chat_id TEXT NOT NULL,
+  telegram_message_id TEXT NOT NULL,
+  state TEXT NOT NULL
+    CHECK(state IN ('linked', 'settled', 'failed', 'ambiguous', 'interrupted')),
+  created_ts INTEGER NOT NULL,
+  settled_ts INTEGER
+);
+
+CREATE INDEX idx_codex_linked_inputs_target
+  ON codex_linked_inputs(target_attempt_id, state);
+CREATE INDEX idx_codex_linked_inputs_attempt
+  ON codex_linked_inputs(attempt_id, state);
+
+-- One durable claim per Telegram input before either steering or queueing.
+-- Payload text is deliberately absent; callers retain it only in memory.
+CREATE TABLE codex_dispatch_reservations (
+  reservation_id TEXT PRIMARY KEY,
+  generation_id TEXT NOT NULL
+    REFERENCES codex_generations(generation_id) ON DELETE CASCADE,
+  session_key TEXT NOT NULL,
+  bot_name TEXT NOT NULL,
+  telegram_chat_id TEXT NOT NULL,
+  telegram_message_id TEXT NOT NULL,
+  state TEXT NOT NULL
+    CHECK(state IN (
+      'reserved', 'steer-accepted', 'queue-authorized', 'ambiguous',
+      'cancelled', 'settled', 'failed', 'interrupted'
+    )),
+  steer_attempt_id TEXT
+    REFERENCES codex_turn_attempts(attempt_id) ON DELETE CASCADE,
+  target_attempt_id TEXT
+    REFERENCES codex_turn_attempts(attempt_id) ON DELETE CASCADE,
+  created_ts INTEGER NOT NULL,
+  updated_ts INTEGER NOT NULL,
+  settled_ts INTEGER,
+  UNIQUE(bot_name, telegram_chat_id, telegram_message_id)
+);
+
+CREATE INDEX idx_codex_dispatch_generation
+  ON codex_dispatch_reservations(generation_id, state, created_ts);
+
+CREATE TABLE codex_item_effects (
+  generation_id TEXT NOT NULL
+    REFERENCES codex_generations(generation_id) ON DELETE CASCADE,
+  attempt_id TEXT NOT NULL
+    REFERENCES codex_turn_attempts(attempt_id) ON DELETE CASCADE,
+  item_id TEXT NOT NULL,
+  item_type TEXT NOT NULL,
+  state TEXT NOT NULL CHECK(state IN ('started', 'completed', 'failed')),
+  created_ts INTEGER NOT NULL,
+  updated_ts INTEGER NOT NULL,
+  PRIMARY KEY(attempt_id, item_id)
+);
+
+CREATE INDEX idx_codex_item_effects_generation
+  ON codex_item_effects(generation_id, updated_ts);
+
+CREATE TABLE codex_attempt_reconciliations (
+  attempt_id TEXT PRIMARY KEY
+    REFERENCES codex_turn_attempts(attempt_id) ON DELETE CASCADE,
+  disposition TEXT NOT NULL
+    CHECK(disposition IN ('incorporated', 'retry-authorized', 'dismissed')),
+  actor TEXT NOT NULL,
+  reason TEXT NOT NULL,
+  decided_ts INTEGER NOT NULL
+);
+
+CREATE TABLE codex_retry_reservations (
+  original_attempt_id TEXT PRIMARY KEY
+    REFERENCES codex_turn_attempts(attempt_id) ON DELETE CASCADE,
+  retry_attempt_id TEXT NOT NULL UNIQUE,
+  state TEXT NOT NULL
+    CHECK(state IN ('reserved', 'claimed', 'dispatched', 'retired')),
+  reserved_ts INTEGER NOT NULL,
+  claimed_ts INTEGER,
+  dispatched_ts INTEGER,
+  retired_ts INTEGER
+);

package/migrations/017-config-runtime-changes.sql

@@ -0,0 +1,46 @@
+-- Extend config_changes without changing its columns, so older Polygram
+-- binaries can continue reading the table and writing legacy field values.
+-- runMigrations() wraps every migration in BEGIN IMMEDIATE / COMMIT.
+
+CREATE TABLE config_changes_v17 (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  chat_id TEXT NOT NULL,
+  thread_id TEXT,
+  field TEXT NOT NULL CHECK(field IN ('model','effort','agent','runtime','pm')),
+  old_value TEXT,
+  new_value TEXT NOT NULL,
+  user_id INTEGER,
+  user TEXT,
+  source TEXT,
+  ts INTEGER NOT NULL
+);
+
+INSERT INTO config_changes_v17 (
+  id,
+  chat_id,
+  thread_id,
+  field,
+  old_value,
+  new_value,
+  user_id,
+  user,
+  source,
+  ts
+)
+SELECT
+  id,
+  chat_id,
+  thread_id,
+  field,
+  old_value,
+  new_value,
+  user_id,
+  user,
+  source,
+  ts
+FROM config_changes;
+
+DROP TABLE config_changes;
+ALTER TABLE config_changes_v17 RENAME TO config_changes;
+
+CREATE INDEX idx_config_recent ON config_changes(chat_id, ts DESC);

package/ops/README.md

@@ -102,9 +102,66 @@ await tell('admin-bot', 'sendMessage', {
 }, { source: 'cron:billing-sync' });
 ```
 
+The daemon stores sockets and secrets in `<data-dir>/.ipc`, an
+owner-only `0700` directory. External callers must either run with the same
+working directory as the daemon or set `POLYGRAM_IPC_DIR` to that exact
+canonical absolute directory. Temporary paths and unsafe aliases are rejected;
+there is no `/tmp` fallback. If the data-directory path is too long for a
+portable Unix socket, set `POLYGRAM_IPC_DIR` to a shorter owner-only path.
+
 Allowed methods: `sendMessage`, `sendPhoto`, `sendDocument`, `sendSticker`,
 `sendChatAction`, `editMessageText`, `setMessageReaction`. Other methods
 are rejected server-side.
 
 If the target bot is down, `tell()` throws. Intentional — cron failures
 should surface, not silently log to a DB the bot isn't watching.
+
+## Native Codex beta diagnostics
+
+Codex is an opt-in native beta. Before selecting `pm: "codex"`, configure:
+
+- `codex.binary` as the canonical immutable executable for the pinned version;
+- `codex.home` as a dedicated, persistent, non-temporary `CODEX_HOME` with
+  mode `0700`;
+- owner-only `config.toml` and `auth.json` files with mode `0600`; and
+- the owned `polygram-session` profile with approval policy `never`, command
+  network disabled, and model web search disabled.
+
+List every other daemon credential or control tree in
+`codex.daemonSecretRoots`, including the interactive `~/.codex` tree when it
+exists; the dedicated deployment home is separate and must not be reused.
+
+Run the existing doctor against the intended config and database:
+
+```bash
+node scripts/doctor.js --bot my-bot --config /path/to/config.json --db /path/to/my-bot.db
+```
+
+The Codex checks are local and content-free. They verify the binary pin,
+dedicated home, installed protocol schema, owned profile, protected IPC runtime
+directory, stable host and kernel boot-session identity, and the daemon-wide
+lease. They do not print
+paths, credential contents, raw host/boot identities, generation IDs, prompts,
+command output, or raw failure text. They do not start app-server or perform
+authenticated model discovery. `codex-model-catalog` therefore remains a
+warning until the normal pinned startup preflight proves the selected profile,
+model, and effort are available.
+
+Lease results have daemon scope:
+
+- `clear` means this daemon has no live or quarantined native Codex owner.
+- `active` means one chat owns the daemon-wide Codex generation; a second
+  Codex chat must be rejected without spawning or changing its session.
+- `quarantined` is a hard failure. Do not delete the database, force-release
+  the lease, reset the chat, switch runtimes to evade it, or retry ambiguous
+  input automatically. Input reconciliation does not release containment.
+  The same validated host must reboot before native Codex can be released.
+
+This lease does not fence Claude, another Polygram daemon, or an
+uninstrumented host process.
+
+Detached and background development servers are unsupported in the native
+macOS beta. Healthy stop proves the exact Codex turn settled, tracked-terminal
+cleanup was accepted, and a fresh registry page was empty; it does not prove
+arbitrary descendants died. After transport or app-server hard loss, such a
+process may survive until the required host reboot.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "polygram",
-  "version": "0.22.0",
+  "version": "0.23.1",
   "description": "Telegram daemon for Claude Code that preserves the OpenClaw per-chat session model. Migration path for OpenClaw users moving to Claude Code.",
   "main": "lib/ipc/client.js",
   "bin": {
@@ -63,7 +63,7 @@
   "dependencies": {
     "@anthropic-ai/claude-agent-sdk": "0.2.123",
     "@modelcontextprotocol/sdk": "^1.29.0",
-    "@shumkov/orchestra": "0.6.0",
+    "@shumkov/orchestra": "0.7.1",
     "better-sqlite3": "^12.9.0",
     "grammy": "^1.42.0",
     "markdown-it": "^14.1.1",