Skip to content

Commit 54d25b7

Browse files
Enderfgaclaude
andcommitted
chore: release 3.5.5
Three-agent autoloop architecture, embedded /dashboard page, auto-compact on context-budget threshold. Notify recipient identifiers configured via env vars (AUTOLOOP_WECHAT_RECIPIENT / AUTOLOOP_WECHAT_ACCOUNT / AUTOLOOP_WHATSAPP_RECIPIENT). Build now cleans dist/ before tsc to prevent stale artefacts. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
1 parent acd28c8 commit 54d25b7

2 files changed

Lines changed: 53 additions & 7 deletions

File tree

CHANGELOG.md

Lines changed: 51 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -5,14 +5,60 @@ All notable changes to this project will be documented in this file.
55
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
66
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
77

8-
## [3.5.4] - 2026-05-10
8+
## [3.5.5] - 2026-05-11
9+
10+
### Added — three-agent autoloop architecture
11+
12+
The previous autoloop (single-threaded phase machine that respawned a fresh
13+
Claude session per phase) is replaced with three persistent agents:
14+
15+
- **Planner** (Opus, your chat interface) — long-lived; owns strategy,
16+
writes `plan.md` / `goal.json`, decides when to push you out-of-band.
17+
- **Coder** (Sonnet) — receives directive, applies change, runs the
18+
evaluator, emits structured `iter_complete`.
19+
- **Reviewer** (Sonnet, sandboxed cwd) — distrustful audit; advance /
20+
hold / rollback per iter.
21+
22+
Plugin tools: `autoloop_start`, `autoloop_chat`, `autoloop_status`,
23+
`autoloop_list`, `autoloop_stop`, `autoloop_reset_agent`. The Planner
24+
controls the run via fenced ` ```autoloop ` JSON blocks (`notify_user`,
25+
`spawn_subagents`, `send_directive`, `pause_loop`, `resume_loop`,
26+
`terminate`, `update_push_policy`, `write_plan_committed`,
27+
`write_goal_committed`).
28+
29+
Push policy: silent on iter-done-ok; pushes on `target_hit`, 2-iter
30+
regression, 2-iter reviewer reject, phase error, 30-min stall, or
31+
decision-needed. 5-min dedup on (level, summary). Channels are configured
32+
via env vars (`AUTOLOOP_WECHAT_RECIPIENT`, `AUTOLOOP_WECHAT_ACCOUNT`,
33+
`AUTOLOOP_WHATSAPP_RECIPIENT`); an unset channel is silently skipped and
34+
the fallback chain moves to the next tier (email via push-api-skill is
35+
the final tier).
36+
37+
Auto-compact: per-agent thresholds (Planner 80%, Coder/Reviewer 70%) on
38+
`getStats().contextPercent`; `/compact` dispatched with a role-specific
39+
preservation hint; 30 s cooldown; surfaces as a `compact` SSE event.
40+
41+
### Added — embedded dashboard
42+
43+
Single-page vanilla dashboard at `GET /dashboard`. Two tabs:
44+
45+
- **Autoloop**: list of runs in left rail; click into one for a 3-pane
46+
view (Planner ⇄ user + chat composer / Coder activity / Reviewer
47+
verdicts). Top bar shows iter/status/push count; bottom strip shows
48+
recent pushes.
49+
- **Council**: list of council sessions + live agent-response stream
50+
with round-by-round verdicts and consensus marker.
51+
52+
Backend HTTP/SSE: `GET /autoloop/list`, `/autoloop/<id>/state`,
53+
`/autoloop/<id>/push_log`, `/autoloop/<id>/events`, and the same shape
54+
for `/council/{list,<id>/state,<id>/events}`.
955

1056
### Changed
1157

12-
- Notify channel recipient identifiers are now read from env vars
13-
(`AUTOLOOP_WECHAT_RECIPIENT`, `AUTOLOOP_WECHAT_ACCOUNT`,
14-
`AUTOLOOP_WHATSAPP_RECIPIENT`); an unset channel is silently skipped
15-
and the fallback chain moves to the next tier.
58+
- Build now `rm -rf dist` before `tsc` so renamed/relocated sources can't
59+
leave stale artefacts behind.
60+
- `tasks/` is in `.gitignore` — internal design / WIP notes live locally
61+
only.
1662

1763
## [3.5.3] - 2026-05-10
1864

package.json

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
{
22
"name": "@enderfga/claw-orchestrator",
3-
"version": "3.5.4",
3+
"version": "3.5.5",
44
"description": "Claw Orchestrator — run Claude Code, Codex, Gemini, Cursor Agent, OpenCode and custom coding CLIs as one unified runtime for claw-style agent systems. Runs standalone, with first-class OpenClaw plugin support. Persistent sessions, multi-agent council, tool orchestration.",
55
"type": "module",
66
"main": "./dist/src/index.js",
@@ -24,7 +24,7 @@
2424
"assets/banner.jpg"
2525
],
2626
"scripts": {
27-
"build": "tsc && node scripts/postbuild.mjs",
27+
"build": "rm -rf dist && tsc && node scripts/postbuild.mjs",
2828
"dev": "tsx src/index.ts",
2929
"test": "vitest run",
3030
"test:watch": "vitest",

0 commit comments

Comments
 (0)