diff --git a/README.md b/README.md
index f9905f9..e8980a4 100644
--- a/README.md
+++ b/README.md
@@ -10,12 +10,12 @@
- 
- 
- 
- 
+ 
+ 
+ 
+ 
- 
+ 
---
@@ -88,12 +88,16 @@ Native floating overlays: orange-bordered recording panel with pulsing red dot,
Select any text, anywhere. Right-click. Eight AI services system-wide: Proofread, Elevate, Explain, Translate, Reply (with `:tone` syntax), Prompt, Read Aloud, Save. Powered by the local LLM.
-### 4. CODEC Chat — 250K Context + 12 Agent Crews
+### 4. CODEC Chat — 250K Context + 12 Agent Crews + Drop-a-Project Autonomy
Full conversational AI. Long context. File uploads (drag-and-drop). Image analysis via vision model. Web search. Persistent conversation history with sidebar. Edit and re-send messages. Regenerate responses.
Voice input via continuous microphone with stop button. Streaming responses with typing and thinking indicators.
+**Four modes** in the chat composer: **Chat / Think / Agents / Project**.
+
+**Project mode (Phase 3 — drop-a-project autonomy).** Type a project description, hit send. CODEC drafts a structured plan via local Qwen-3.6 with an explicit permission manifest (skills, write paths, network domains, destructive ops). You approve once. `codec-agent-runner` (PM2 daemon) picks up the approved plan and executes it autonomously — Qwen ↔ skill loops, permission-gated, plan-hash tamper detection, resume-after-restart, multi-agent concurrency capped at 3. Status pills above the input show running/paused/blocked agents with inline approve/pause/resume/abort actions. Updates post back into the chat thread. Try: *"Build me a Telegram bot that monitors Marbella property listings under €500k"* or *"Watch EUR/USD intraday vol and ping me when 30-min realized vol crosses 0.4%"*.
+
Plus 12 autonomous agent crews — not single prompts, full multi-step workflows. Say *"research the latest AI agent frameworks and write a report."* Minutes later there's a formatted Google Doc in Drive with sources, images, and recommendations.
| Crew | Output |
@@ -115,6 +119,8 @@ Schedule any crew: *"Run competitor analysis every Monday at 9am"*
The multi-agent framework is under 800 lines. Zero dependencies. No CrewAI. No LangChain. 7 built-in tools including web search, file operations, Google Docs, image generation, and vision.
+**Phase 3 substrate** (autonomous agents) reuses Phase 1+2 components: unified audit envelope (Step 1), plugin lifecycle hooks (Step 2), AskUser + strict-consent + step budget (Step 3), continuous observation loop (Step 5), trigger system (Step 6), end-of-day shift report (Step 7). Per-agent state at `~/.codec/agents//`. Global allowlist tier at `~/.codec/agent_global_grants.json`. 17 new audit events. See `docs/PHASE3-COMPLETE.md` for the full sign-off.
+
### 5. CODEC Vibe — AI Coding IDE + Skill Forge
Split-screen in the browser. Monaco editor on the left (same engine as VS Code, v0.45.0). AI chat on the right. Describe what's needed — CODEC writes it, click Apply, run it, live preview in browser.
@@ -604,6 +610,16 @@ python3 setup_codec.py
## What's Coming
+**Phase 3.5 (in progress)** — UX + polish on top of the autonomous-agent substrate:
+
+- [ ] **Anchor example end-to-end run** — drop the Marbella property bot project, document the run from plan-draft to final notification (`docs/PHASE35-ANCHOR-EXAMPLE-RUN.md`)
+- [ ] **Proactive intelligence overlay** — observer-driven contextual nudges via Step 6 trigger system (*"You've been on this Notion doc 30 min, want a summary?"*) with strict not-invasive defaults (1 suggestion / hour max, easy dismiss, per-pattern kill switch)
+- [ ] **Dedicated `blocked_on_qwen` status** — daemon-driven auto-resume when Qwen recovers (cleaner UX than reusing `blocked_on_permission` for LLM outages — Step 9 review C2)
+- [ ] **Read-paths runtime enforcement** — `Action.reads_path` field + LLM prompt update to symmetric read/write gating (Step 9 review M4)
+- [ ] **Multi-channel notifications** — at agent-spawn time, optionally route updates to macOS banner / iMessage / Telegram alongside PWA
+
+**Beyond Phase 3:**
+
- [ ] iMessage voice notes (TTS briefing delivered as audio attachment)
- [ ] WhatsApp integration (third messaging channel)
- [ ] Linux support
diff --git a/tests/test_agent_messaging.py b/tests/test_agent_messaging.py
index 130fb7c..d43cca3 100644
--- a/tests/test_agent_messaging.py
+++ b/tests/test_agent_messaging.py
@@ -65,6 +65,13 @@ def temp_codec_dir(tmp_path, monkeypatch):
monkeypatch.setattr(cam, "_CODEC_DIR", tmp_path)
monkeypatch.setattr(cam, "_AGENTS_DIR", tmp_path / "agents")
monkeypatch.setattr(cam, "_NOTIFICATIONS_PATH", tmp_path / "notifications.json")
+ # Redirect codec_audit._AUDIT_LOG so post_message audit emits don't
+ # leak into the production ~/.codec/audit.log (test pollution fix).
+ try:
+ import codec_audit
+ monkeypatch.setattr(codec_audit, "_AUDIT_LOG", tmp_path / "audit.log")
+ except Exception:
+ pass
# Also patch codec_agent_plan paths so _run_agent tests don't touch real ~/.codec
try:
import codec_agent_plan as cap
diff --git a/tests/test_agent_plan.py b/tests/test_agent_plan.py
index 6a35407..4fcfe0c 100644
--- a/tests/test_agent_plan.py
+++ b/tests/test_agent_plan.py
@@ -117,9 +117,13 @@ def test_plan_hash_changes_when_content_changes():
@pytest.fixture
def temp_codec_dir(tmp_path, monkeypatch):
import codec_agent_plan as cap
+ import codec_audit
monkeypatch.setattr(cap, "_CODEC_DIR", tmp_path)
monkeypatch.setattr(cap, "_AGENTS_DIR", tmp_path / "agents")
monkeypatch.setattr(cap, "_GLOBAL_GRANTS_PATH", tmp_path / "agent_global_grants.json")
+ # Test-pollution fix: redirect codec_audit._AUDIT_LOG so audit emits
+ # during plan creation/approval do NOT leak into production ~/.codec/audit.log.
+ monkeypatch.setattr(codec_audit, "_AUDIT_LOG", tmp_path / "audit.log")
return tmp_path
diff --git a/tests/test_agent_runner.py b/tests/test_agent_runner.py
index 52a986e..1c7ba64 100644
--- a/tests/test_agent_runner.py
+++ b/tests/test_agent_runner.py
@@ -245,11 +245,18 @@ def test_destructive_consent_timeout_overnight(monkeypatch):
@pytest.fixture
def temp_codec_dir(tmp_path, monkeypatch):
- """Mirrors Step 8 fixture; redirects all codec_agent_plan paths to tmp."""
+ """Mirrors Step 8 fixture; redirects all codec_agent_plan paths to tmp.
+
+ Also redirects codec_audit._AUDIT_LOG so test runs do NOT pollute
+ the production ~/.codec/audit.log. Without this, _run_agent's audit
+ emits during tests leak into the live log (same pattern as the
+ 2026-05-01 incident; same fix)."""
import codec_agent_plan as cap
+ import codec_audit
monkeypatch.setattr(cap, "_CODEC_DIR", tmp_path)
monkeypatch.setattr(cap, "_AGENTS_DIR", tmp_path / "agents")
monkeypatch.setattr(cap, "_GLOBAL_GRANTS_PATH", tmp_path / "agent_global_grants.json")
+ monkeypatch.setattr(codec_audit, "_AUDIT_LOG", tmp_path / "audit.log")
return tmp_path