v0.94: history retention, saved-history access, universal web-agent observation

VERSION

@@ -1 +1 @@
-0.93.0
+0.94.0

browser_extension/agent_surface_adapters.js

@@ -0,0 +1,98 @@
+(() => {
+  'use strict';
+
+  const PROVIDERS = [
+    {key:'chatgpt', name:'ChatGPT', hosts:['chatgpt.com','chat.openai.com']},
+    {key:'claude', name:'Claude', hosts:['claude.ai']},
+    {key:'microsoft_copilot', name:'Microsoft Copilot', hosts:['copilot.microsoft.com','copilot.microsoft365.com','m365.cloud.microsoft']},
+    {key:'lovable', name:'Lovable', hosts:['lovable.dev']},
+    {key:'gemini', name:'Gemini', hosts:['gemini.google.com']},
+  ];
+  const STOP_RE=/^(stop|cancel)( generating| response| task| run)?$/i;
+  const SEND_RE=/^(send|submit|ask|run|build|generate)( prompt| message| request)?$/i;
+  const APPROVE_RE=/^(allow|approve|confirm|continue|accept)$/i;
+
+  function providerForHost(host){
+    const h=String(host||'').toLowerCase().replace(/^www\./,'');
+    return PROVIDERS.find(p=>p.hosts.some(x=>h===x||h.endsWith('.'+x)))||null;
+  }
+  function accessibleName(el){
+    if(!el)return '';
+    return String(el.getAttribute?.('aria-label')||el.getAttribute?.('title')||el.textContent||'').replace(/\s+/g,' ').trim().slice(0,80);
+  }
+  function buttonLike(el){return !!el?.closest?.('button,[role="button"],input[type="submit"]');}
+  function hasBusyState(doc){
+    if(!doc)return false;
+    if(doc.querySelector('[aria-busy="true"]'))return true;
+    return [...doc.querySelectorAll('button,[role="button"]')].some(el=>STOP_RE.test(accessibleName(el)));
+  }
+  function hasVisibleErrorState(doc){return !!doc?.querySelector?.('[role="alert"][aria-live], [role="alert"]');}
+  function hasApprovalState(doc){
+    if(!doc)return false;
+    for(const root of doc.querySelectorAll('dialog,[role="dialog"]')){
+      if([...root.querySelectorAll('button,[role="button"]')].some(el=>APPROVE_RE.test(accessibleName(el))))return true;
+    }
+    return false;
+  }
+  function isSendControl(el){const control=buttonLike(el);return !!control&&SEND_RE.test(accessibleName(control));}
+  function isStopControl(el){const control=buttonLike(el);return !!control&&STOP_RE.test(accessibleName(control));}
+  function isApprovalControl(el){const control=buttonLike(el);return !!control&&APPROVE_RE.test(accessibleName(control))&&!!control.closest('dialog,[role="dialog"]');}
+  function safeRunId(){
+    try{return 'web-'+crypto.randomUUID().replaceAll('-','');}catch(_){return 'web-'+Date.now().toString(36)+Math.random().toString(36).slice(2,12);}
+  }
+  function structuralPayload(provider, action, runId, source){
+    return {
+      type:'workflow_observer_event',
+      observed_at:new Date().toISOString(),
+      action,
+      // background.js derives the real safe page URL from sender.tab. Deliberately
+      // do not send conversation paths, titles, prompts, responses or alert text.
+      page:{url:location.origin+'/',title:provider.name},
+      target:{role:'agent-lifecycle',label:provider.name},
+      metadata:{agent_provider:provider.key,agent_run_id:runId,state_source:String(source||'semantic_state').slice(0,40),top_frame:true}
+    };
+  }
+  function sendLifecycle(provider,action,runId,source){
+    // No DOM text or user/model content is included in this message.
+    const message=structuralPayload(provider,action,runId,source);
+    try{
+      const runtime=(typeof browser!=='undefined'&&browser.runtime)?browser.runtime:(typeof chrome!=='undefined'&&chrome.runtime?chrome.runtime:null);
+      runtime?.sendMessage(message).catch?.(()=>{});
+    }catch(_){}
+  }
+
+  function start(doc=typeof document!=='undefined'?document:null){
+    if(!doc||typeof location==='undefined')return;
+    const provider=providerForHost(location.hostname);if(!provider)return;
+    let runId='',wasBusy=false,errorSent=false,approvalSent=false,finishTimer=null;
+    const begin=source=>{if(runId)return;runId=safeRunId();wasBusy=false;errorSent=false;approvalSent=false;sendLifecycle(provider,'agent_run_started',runId,source);};
+    const reset=()=>{runId='';wasBusy=false;errorSent=false;approvalSent=false;if(finishTimer){clearTimeout(finishTimer);finishTimer=null;}};
+    const finish=(action='agent_run_finished',source='busy_cleared')=>{if(!runId)return;sendLifecycle(provider,action,runId,source);reset();};
+    const sample=()=>{
+      const busy=hasBusyState(doc);
+      if(busy&&!runId)begin('busy_state');
+      if(runId){
+        if(busy)wasBusy=true;
+        if(hasVisibleErrorState(doc)&&!errorSent){sendLifecycle(provider,'agent_error',runId,'aria_alert_present');errorSent=true;}
+        if(hasApprovalState(doc)&&!approvalSent){sendLifecycle(provider,'agent_approval_requested',runId,'semantic_dialog');approvalSent=true;}
+        if(wasBusy&&!busy&&!finishTimer){finishTimer=setTimeout(()=>{finishTimer=null;if(runId&&!hasBusyState(doc))finish();},900);}
+        if(busy&&finishTimer){clearTimeout(finishTimer);finishTimer=null;}
+      }
+    };
+    doc.addEventListener('click',ev=>{
+      if(isSendControl(ev.target)){begin('send_control');setTimeout(sample,0);return;}
+      if(runId&&isStopControl(ev.target)){finish('agent_run_cancelled','stop_control');return;}
+      if(runId&&isApprovalControl(ev.target)){sendLifecycle(provider,'agent_approval_received',runId,'approval_control');approvalSent=true;}
+    },true);
+    doc.addEventListener('submit',()=>{begin('form_submit');setTimeout(sample,0);},true);
+    const observer=new MutationObserver(()=>sample());
+    const root=doc.documentElement||doc;observer.observe(root,{subtree:true,childList:true,attributes:true,attributeFilter:['aria-busy','aria-live','role','open','disabled']});
+    setInterval(()=>{if(!doc.hidden)sample();},1500);
+    sample();
+  }
+
+  globalThis.__OWG_AGENT_SURFACE_ADAPTERS_FOR_TESTS__={PROVIDERS,providerForHost,accessibleName,hasBusyState,hasVisibleErrorState,hasApprovalState,isSendControl,isStopControl,isApprovalControl,structuralPayload};
+  if(typeof document!=='undefined'&&typeof MutationObserver!=='undefined'){
+    if(document.readyState==='loading')document.addEventListener('DOMContentLoaded',()=>start(),{once:true});else start();
+  }
+})();

browser_extension/manifest.json

@@ -1,48 +1,22 @@
 {
   "manifest_version": 3,
   "name": "Workflow Observer Browser Sensor",
-  "version": "1.11.0",
-  "version_name": "1.11.0-v58-metadata-signals",
-  "description": "Local-only semantic browser telemetry for Workflow Observer. Query values/fragments, typed field values, clipboard contents, filenames, file contents, and ordinary key identities are not collected.",
+  "version": "1.12.0",
+  "version_name": "1.12.0-v94-agent-lifecycle",
+  "description": "Local-only structural browser telemetry for OpenWorkGraph. Query values/fragments, typed field values, clipboard contents, filenames, file contents, prompts and model responses are not collected.",
   "incognito": "not_allowed",
-  "permissions": [
-    "tabs",
-    "webNavigation",
-    "storage",
-    "alarms"
-  ],
-  "host_permissions": [
-    "http://127.0.0.1:8787/*",
-    "http://*/*",
-    "https://*/*"
-  ],
+  "permissions": ["tabs", "webNavigation", "storage", "alarms"],
+  "host_permissions": ["http://127.0.0.1:8787/*", "http://*/*", "https://*/*"],
   "background": {
-    "scripts": [
-      "browser_auth.js",
-      "background.js",
-      "workflow_enrichment.js",
-      "browser_signal_enrichment.js"
-    ],
+    "scripts": ["browser_auth.js", "background.js", "workflow_enrichment.js", "browser_signal_enrichment.js"],
     "service_worker": "secure_background.js"
   },
-  "content_scripts": [
-    {
-      "matches": [
-        "http://*/*",
-        "https://*/*"
-      ],
-      "js": [
-        "content.js",
-        "content_enrichment.js",
-        "browser_signals.js"
-      ],
-      "run_at": "document_start",
-      "all_frames": true,
-      "match_about_blank": true
-    }
-  ],
-  "action": {
-    "default_title": "Workflow Observer Browser Sensor",
-    "default_popup": "popup.html"
-  }
+  "content_scripts": [{
+    "matches": ["http://*/*", "https://*/*"],
+    "js": ["content.js", "content_enrichment.js", "browser_signals.js", "agent_surface_adapters.js"],
+    "run_at": "document_start",
+    "all_frames": true,
+    "match_about_blank": true
+  }],
+  "action": {"default_title": "Workflow Observer Browser Sensor", "default_popup": "popup.html"}
 }

collector/outbox.py

@@ -8,6 +8,7 @@
 
 from sensitive_identifiers import sanitize_event_identifiers
 from shared.evidence_deletion import event_overlaps_range
+from shared.history_policy import event_kind
 
 
 class EventOutbox:
@@ -56,8 +57,6 @@ def _sanitize_existing_pending(self) -> None:
                     safe = sanitize_event_identifiers(event)
                     payload = json.dumps(safe, ensure_ascii=False, separators=(",", ":"))
                 except Exception:
-                    # Do not destroy a delivery queue merely because one legacy
-                    # payload is malformed; new events are always sanitized.
                     continue
                 if payload != row["payload_json"]:
                     conn.execute(
@@ -93,6 +92,18 @@ def acknowledge(self, event_ids: list[str]) -> None:
         with self._lock, self._connect() as conn:
             conn.execute(f"DELETE FROM pending_events WHERE event_id IN ({placeholders})", tuple(ids))
 
+    def _delete_ids(self, conn: sqlite3.Connection, delete_ids: list[str]) -> int:
+        if not delete_ids:
+            return 0
+        for offset in range(0, len(delete_ids), 500):
+            chunk = delete_ids[offset : offset + 500]
+            placeholders = ",".join("?" for _ in chunk)
+            conn.execute(
+                f"DELETE FROM pending_events WHERE event_id IN ({placeholders})",
+                tuple(chunk),
+            )
+        return len(delete_ids)
+
     def prune_range(self, since: str, until: str) -> int:
         """Remove queued evidence that overlaps a durable local deletion range."""
         with self._lock, self._connect() as conn:
@@ -105,16 +116,27 @@ def prune_range(self, since: str, until: str) -> int:
                     continue
                 if isinstance(event, dict) and event_overlaps_range(event, since, until):
                     delete_ids.append(str(row["event_id"]))
-            if not delete_ids:
-                return 0
-            for offset in range(0, len(delete_ids), 500):
-                chunk = delete_ids[offset : offset + 500]
-                placeholders = ",".join("?" for _ in chunk)
-                conn.execute(
-                    f"DELETE FROM pending_events WHERE event_id IN ({placeholders})",
-                    tuple(chunk),
-                )
-            return len(delete_ids)
+            return self._delete_ids(conn, delete_ids)
+
+    def prune_sessions(self, kind: str, session_ids: list[str]) -> int:
+        """Remove queued rows belonging to expired human or agent sessions."""
+        selected_kind = "agent" if str(kind).lower() == "agent" else "human"
+        wanted = {str(value) for value in session_ids if str(value)}
+        if not wanted:
+            return 0
+        with self._lock, self._connect() as conn:
+            rows = conn.execute("SELECT event_id, payload_json FROM pending_events").fetchall()
+            delete_ids: list[str] = []
+            for row in rows:
+                try:
+                    event = json.loads(row["payload_json"])
+                except Exception:
+                    continue
+                if not isinstance(event, dict):
+                    continue
+                if str(event.get("session_id") or "") in wanted and event_kind(event) == selected_kind:
+                    delete_ids.append(str(row["event_id"]))
+            return self._delete_ids(conn, delete_ids)
 
     def mark_failed(self, event_ids: list[str], error: str) -> None:
         ids = [str(x) for x in event_ids if x]