devops-bot-sdk 1.4.92__tar.gz → 1.4.96__tar.gz

{devops_bot_sdk-1.4.92 → devops_bot_sdk-1.4.96}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: devops-bot-sdk
-Version: 1.4.92
+Version: 1.4.96
 Summary: DevOps Bot Desktop SDK — thin client for the AgentOS Electron desktop app
 Author: noumanaziz2128
 License-Expression: LicenseRef-Proprietary

{devops_bot_sdk-1.4.92 → devops_bot_sdk-1.4.96}/devops_bot_sdk.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: devops-bot-sdk
-Version: 1.4.92
+Version: 1.4.96
 Summary: DevOps Bot Desktop SDK — thin client for the AgentOS Electron desktop app
 Author: noumanaziz2128
 License-Expression: LicenseRef-Proprietary

{devops_bot_sdk-1.4.92 → devops_bot_sdk-1.4.96}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "devops-bot-sdk"
-version = "1.4.92"
+version = "1.4.96"
 description = "DevOps Bot Desktop SDK — thin client for the AgentOS Electron desktop app"
 readme = "README.md"
 license = "LicenseRef-Proprietary"

{devops_bot_sdk-1.4.92 → devops_bot_sdk-1.4.96}/sdk/__init__.py

@@ -1,6 +1,6 @@
 """AgentOS Desktop SDK — thin HTTPS/SSE client for the Electron app.
 
-Version: 1.4.92
+Version: 1.4.96
 
 Public surface:
   BackendClient.from_config()        — create client from ~/.agentos/config.toml
@@ -30,7 +30,7 @@ Rules:
   - All data egress through submit_webhook only
 """
 
-__version__ = "1.4.92"
+__version__ = "1.4.96"
 __author__  = "AgentOS"
 
 from sdk.client import BackendClient

{devops_bot_sdk-1.4.92 → devops_bot_sdk-1.4.96}/sdk/client.py

@@ -58,7 +58,7 @@ def _notify_status(task_id: str, status: str) -> None:
             pass
 
 
-SDK_VERSION    = "1.4.92"
+SDK_VERSION    = "1.4.96"
 _POLL_INTERVAL = 3.0
 _POLL_TIMEOUT  = 600.0
 _ORCHESTRATE_TIMEOUT      = 2700.0  # 45 min — covers approval wait + VPS execution time
@@ -849,9 +849,17 @@ class BackendClient:
         headers = {k: v for k, v in self._headers.items() if k.lower() != "content-type"}
         headers.update({"accept": "*/*", "developer-name": "null", "product": ""})
         # POST /tasks-ml must never carry an SLA target — the backend validator
-        # rejects any value not shaped like "8h", and SLA is not ours to set.
+        # rejects any value not shaped like "8h" (an empty "" 400s with
+        # "Slatarget must be a positive integer of hours…"), and SLA is not ours
+        # to set. Drop it by NORMALISED name so every casing/separator variant
+        # (slaTarget, slaTargets, sla_target, SLATarget, sla-target, …) is
+        # excluded — never sent as "" or anything else, always an empty field.
+        def _is_sla(key: str) -> bool:
+            norm = "".join(ch for ch in key.lower() if ch.isalnum())
+            return norm in ("slatarget", "slatargets")
+
         files = [(k, (None, str(v))) for k, v in fields.items()
-                 if v is not None and k not in ("slaTarget", "slaTargets")]
+                 if v is not None and not _is_sla(k)]
         for m in (model_used or []):
             files.append(("modelUsed[]", (None, str(m))))  # [] → backend parses as array
         try:

{devops_bot_sdk-1.4.92 → devops_bot_sdk-1.4.96}/sdk/ipc/handlers.py

@@ -416,7 +416,10 @@ async def _run_with_heartbeat(coro, *, send: Callable, thread_id: str, label: st
         beat.cancel()
         try:
             await beat
-        except Exception:  # noqa: BLE001
+        except (asyncio.CancelledError, Exception):  # noqa: BLE001
+            # CancelledError is a BaseException: raised here when the beat task
+            # was cancelled before it ever ran (coro finished immediately) — it
+            # must not propagate and sink the caller's result.
             pass
 
 
@@ -441,6 +444,7 @@ async def _run_generation(
         send=send, thread_id=thread_id, label=label, name=name,
     )
     combined = res.get("result") or ""
+    longest = res.get("longest_text") or ""
     sid = res.get("session_id")
     chunks = 0
     while res.get("output_cap_exceeded") and sid and chunks < _MAX_OUTPUT_CAP_RESUMES:
@@ -463,9 +467,24 @@ async def _run_generation(
             send=send, thread_id=thread_id, label=f"{label} (cont.)", name=name,
         )
         combined += res.get("result") or ""
+        if len(res.get("longest_text") or "") > len(longest):
+            longest = res.get("longest_text") or ""
         sid = res.get("session_id") or sid
     out = dict(res)
     out["result"] = combined      # the full, reassembled output
+    # Salvage a lost artifact: the CLI's `result` is only the LAST assistant
+    # message. If the model wrote the artifact and was then woken once more by a
+    # stray late event (e.g. a leftover background timer) for a short closing
+    # remark ("nothing further is pending"), `combined` is that remark and the
+    # real output is an EARLIER, much longer message. When an earlier message
+    # dwarfs the final one, it is the artifact — use it.
+    if (not out.get("limit_exceeded") and len(longest) > 1000
+            and len(longest) > 2 * len(combined.strip())):
+        logger.warning(
+            "generation.salvaged_longest_message final_len=%s longest_len=%s",
+            len(combined.strip()), len(longest),
+        )
+        out["result"] = longest.strip()
     out["session_id"] = sid
     return out
 
@@ -1336,13 +1355,22 @@ async def _handle_story(
         # The Story is complete — nothing to convert or code in this run.
         return None
 
-    # STEP 1 — validate whether this is a CODEBASE (development) task: repo links /
-    # file paths in the instructions, or dev signals (PRs, branches, endpoints,
-    # deploy, …). A codebase task ALWAYS gets a technical PRD built from the current
-    # codebase; other domains (Finance, HR, …) get one only when explicitly asked.
+    # STEP 1 — do we have CODEBASE UNDERSTANDING to ground a PRD in? That means an
+    # actual repo target (a github/gitlab/bitbucket link in the instructions OR the
+    # story), or dev work called for in the OPERATOR INSTRUCTIONS (PRs, branches,
+    # endpoints, deploy, …).
+    #
+    # Deliberately NOT keyed off the story DESCRIPTION's vocabulary or a "PRD"
+    # mention: a business/BA user story that merely contains technical words
+    # (endpoint, API, database, "PRD", …) is NOT a codebase task. Keying off the
+    # description used to false-positive such stories into PRD generation, which
+    # folded an unwanted PRD into — and mangled — the operator's own document.
     has_repo = bool(_extract_repo_targets(human)) or bool(_extract_repo_targets(user_story))
-    is_codebase_task = has_repo or _is_dev_task(human) or _is_dev_task(user_story)
-    wants_prd = is_codebase_task or _wants_prd(human) or _wants_prd(user_story)
+    is_codebase_task = has_repo or _is_dev_task(human)
+    # PRD ONLY when we have codebase understanding. Without a repo/codebase, the
+    # Story is converted to a Task with its description left intact (the whole,
+    # proper document the operator provided) — never rewritten with a PRD.
+    wants_prd = is_codebase_task
 
     if is_codebase_task:
         step = ("validated as a codebase task — building a technical PRD from the "
@@ -1382,6 +1410,13 @@ async def _handle_story(
         # Use the output regardless of an is_error flag (a stray read-only tool denial
         # shouldn't discard a PRD the model actually produced); only a usage limit is fatal.
         prd = "" if gen.get("limit_exceeded") else (gen.get("result") or "").strip()
+        if prd and len(prd) < 400:
+            # A real PRD (numbered sections per _PRD_FORMAT) is never this short —
+            # this is a stub (e.g. a closing remark captured as the run's result).
+            # Treat it as a failed generation rather than filing garbage in the ticket.
+            if not gen.get("error"):
+                gen["error"] = f"PRD too short ({len(prd)} chars) — not a document"
+            prd = ""
         if prd:
             # PRD lives alongside the user story in the (now Task) description.
             new_description = f"{user_story}\n\n---\n\n{_PRD_MARKER}\n\n{prd}"[:20000]

{devops_bot_sdk-1.4.92 → devops_bot_sdk-1.4.96}/sdk/local_exec.py

@@ -386,6 +386,12 @@ async def run_claude_local(
 
     result_event: dict | None = None
     files_changed: list[str] = []
+    # Longest single assistant TEXT message seen in the stream. The CLI's final
+    # `result` field is only the LAST assistant message — if the model outputs a
+    # large artifact (PRD/decomposition) and is then woken once more by a stray
+    # late event (e.g. a leftover background timer) for a short closing remark,
+    # the artifact would be lost. Callers use this to salvage it.
+    longest_text = {"text": ""}
     session_id: str | None = resume_session_id
     import time as _time
     started = _time.monotonic()
@@ -452,6 +458,14 @@ async def run_claude_local(
             if event.get("session_id"):
                 session_id = str(event["session_id"])
 
+            # Track the longest assistant text message (see `longest_text` above).
+            if event.get("type") == "assistant":
+                for c in event.get("message", {}).get("content", []) or []:
+                    if isinstance(c, dict) and c.get("type") == "text":
+                        txt = str(c.get("text") or "")
+                        if len(txt) > len(longest_text["text"]):
+                            longest_text["text"] = txt
+
             # Track files the agent created/edited for the result summary.
             if event.get("type") == "user":
                 tr = event.get("tool_use_result")
@@ -526,6 +540,7 @@ async def run_claude_local(
     return _finish({
         "ok": (not is_error) and (not limit_hit),
         "result": result_text,
+        "longest_text": longest_text["text"],
         "input_tokens": in_tok,
         "output_tokens": out_tok,
         "total_tokens": in_tok + out_tok,