Skip to content

CHANGELOG.md

Latest commit

 

History

History
56 lines (40 loc) · 7.18 KB

CHANGELOG.md

File metadata and controls

56 lines (40 loc) · 7.18 KB
Raw

Changelog

Unreleased

Added

  • Session degradation watch (bin/iak-degradation-watch.mjs): a per-machine daemon that tails watched agents' Claude Code session transcripts and posts a room alert (pushed to the owner's phone via CodeWatch) the moment a session's message.model leaves the configured accepted set or its permissionMode differs from the expected one. Motivation: on 2026-07-18 an agent's session silently slipped to a small fallback model with manual permission prompts while the owner was away; the degraded agent kept posting — looping on stale tasks and reposting completed instructions — and cost an evening to diagnose. Alerts fire immediately on first detection (a wrong model in a transcript is a fact, not a flap) with a 30-minute per-signal cooldown and a recovery note. Config: degradation_watch block (interval_sec, alert_room, agents[] with transcript_dir, accepted_models prefix list, expected_permission_mode); API key reuses poller.api_key. --once and --dry-run flags for testing. Run one instance per machine under process supervision.
  • Send-to-session: the :8788 daemon gains POST /sessions/send {agent, text, from?} — deliver text as user input into any named agent's live session — plus GET /sessions/agents for the target picker. Four delivery adapters, configured per agent under mcp.confirmations.sessions.agents: wake (spawn a wake/injection script), gui (same, chained behind a human-idle guard with --wait so it never types over the user), tmux (literal send-keys + Enter), and forward (relay to the owning machine's daemon, provenance passed through so the [from …] prefix is applied exactly once). Deliveries are accepted-async (202) and every attempt is receipted. This is the backend half of the CodeWatch send box; it drives any agent uniformly, including headless sessions the provider apps cannot reach.
  • Responder-lock enforcement in the MCP room tools (#42): room_post and alert_recipient now refuse when another live session holds the machine's room-responder lock — a PASSIVE session that ignores its injected instructions still cannot post as the agent through IAK tooling. Lock semantics mirror the SessionStart hook exactly (pid + process-start-time identity, stale locks ignored, fail-open on mechanics); the owning session is recognized by finding the lock's pid in the caller's process-ancestor chain. Read-only room tools (room_recent) are unaffected. Raw HTTP posting against GroupMind is out of scope client-side and tracked in issue #42's server-side follow-up.

Fixed

  • Duplicate room-responder voices: session-bootstrap.sh now enforces a single room responder per notification file via a lock file (<notification file>.responder.lock). The first session claims the lock and gets the full bootstrap; any later session on the same machine gets PASSIVE instructions (serve the user directly, never answer the room) instead of booting a second copy of the same handle. The lock stores the owner's pid + session id + process start time, so the owner reclaims it across resume/compact, a lock whose owner process died is stolen automatically, and a recycled pid never masquerades as the owner (adversarial review by @codexmb: initial claim is O_EXCL-atomic, aliveness uses ps so a cross-user EPERM never reads as dead, and every winner re-verifies ownership after a settle window so concurrent starts elect exactly one responder). Disable with IAK_RESPONDER_LOCK=off, relocate with IAK_RESPONDER_LOCK=<path>. Motivation: on 2026-07-16 a stray claude rc session on the Mac mini booted the same bootstrap and posted to the room as a second @claudemm, producing conflicting answers.

0.9.0 (2026-06-10)

Added

  • user-intent-kit embedded: the UIK repo now lives at packages/user-intent-kit (JS client + adapters + Python/Swift/Kotlin ports) and is wired in as a file: dependency. It remains independently publishable to npm.
  • src/intent.mjs: single construction point for intent clients/adapters from the intent config block (camelCase and snake_case keys both accepted). Includes a 30s intent cache shared by enrichment and nudge gating.
  • intent CLI command: ide-agent-kit intent <get|profile|derived|state|patch|heartbeat|daemon>. state shows the two-level state plus the background gate verdict; daemon is the embedded equivalent of uik-daemon driven by config.intent.* instead of env vars.
  • Nudge suppression in rooms watch: when config.intent is set and the user's urgency_mode is emergency-only, the poller still writes notification files but skips the tmux/command nudge. Opt out with intent.suppress_nudges: false. Fails open on API errors.

Fixed

  • Intent enrichment never worked: enrichment.mjs authenticated with Authorization: Bearer but the intent API expects X-API-Key, so every event carried the placeholder intent and an enrichment error. Now uses the embedded IntentClient (correct auth) plus the shared cache, so message bursts no longer trigger one intent fetch per event.
  • loadConfig dropped the intent, memory_api, and moltbook blocks: it rebuilds the config from an allowlist of known sections, and these three were missing — so intent gating, intent/memory enrichment, and moltbook config silently never received their settings through any CLI entry point. All three now pass through (with a regression test).

Changed

  • fetchIntentGate in background.mjs is built on the embedded IntentClient instead of a hand-rolled fetch. Same gate rules, same return shape, same fail-open behavior, same 5s timeout.
  • npm test now also runs the embedded user-intent-kit test suite.

0.6.1 (2026-04-07)

Docs

  • README: new "Publishing state (the other side)" subsection under User Intent Kit, explaining that reading intent via IAK enrichment is only half the loop. Agents also need to publish their own slot heartbeats via uik-daemon from user-intent-kit >= 0.2.1. Without the daemon slots go stale and other agents treat them as offline.

0.6.0 (2026-04-07)

Added

  • DM inbox polling in the room watcher (merged from #9 codexmb/dm-poller-support). The poller now watches direct messages addressed to the agent handle in addition to room messages, writing them into the same notification file.
  • Background consolidation first cut: light / REM / deep phases with sidecar state under ~/.iak/consolidation/ (f80bd56).
  • UIK intent gating for background consolidation: checks reachability_mode and suppress_audio before scheduling deep phases, so heavy work only runs when the user is reachable (e2be4a6).
  • Config documentation for the dm_poller block and a working config/macbook.json reference.

Fixed

  • config/macbook.json dm_poller.api_key was paired with the wrong handle in an earlier draft — fixed to use the correct @claudeMB key.

Docs

  • README: User Intent Kit section explaining UIK fields and agent usage (3a46eae).
  • README: v0.5.0 memory payload shapes and verification commands (59b8da6).

Compatibility

No breaking changes to config schema. dm_poller is additive; existing configs keep working.

0.5.0

Memory backend hooks, payload schema documentation.

0.4.0

Initial OpenClaw gateway integration.