codee-agent 0.5.0__tar.gz → 0.5.2__tar.gz

{codee_agent-0.5.0 → codee_agent-0.5.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codee-agent
-Version: 0.5.0
+Version: 0.5.2
 Summary: A virtual co-worker that picks up Jira and Azure DevOps tasks and solves them with coding agents.
 Keywords: ai,agent,jira,azure-devops,claude-code,github-copilot,automation
 Author: Denis Kibalko
@@ -35,11 +35,11 @@ Description-Content-Type: text/markdown
 
 # Codee - a virtual co-worker
 
-The goal of this project is to provide an assistant that will integrate into your existing environment in order to help you to solve tasks.
+The goal of this project is to provide an assistant that will integrate into your existing environment in order to help you to work on tasks (development, code review, testing).
 
-Codee works with Jira and Azure DevOps as tasks provider, new providers are quite easy to create, PRs are welcome.
+Codee works with Jira and Azure DevOps as tasks providers, new providers are quite easy to create, PRs are welcome.
 
-As agents it current works with Claude Code and Github Copilot.
+As agents it currently works with Claude Code and Github Copilot.
 
 ## Run Codee
 
@@ -71,6 +71,3 @@ uv run codee-start --debug-all   # ...plus reflex, boto3, urllib3 and friends
 ### Skills
 
 Codee uses skills to perform tasks. It is the same skills which are used in Claude Code, Codex and other agents.
-
-
-TODO: add skill types, x-codee skills extension

{codee_agent-0.5.0 → codee_agent-0.5.2}/README.md

@@ -1,10 +1,10 @@
 # Codee - a virtual co-worker
 
-The goal of this project is to provide an assistant that will integrate into your existing environment in order to help you to solve tasks.
+The goal of this project is to provide an assistant that will integrate into your existing environment in order to help you to work on tasks (development, code review, testing).
 
-Codee works with Jira and Azure DevOps as tasks provider, new providers are quite easy to create, PRs are welcome.
+Codee works with Jira and Azure DevOps as tasks providers, new providers are quite easy to create, PRs are welcome.
 
-As agents it current works with Claude Code and Github Copilot.
+As agents it currently works with Claude Code and Github Copilot.
 
 ## Run Codee
 
@@ -36,6 +36,3 @@ uv run codee-start --debug-all   # ...plus reflex, boto3, urllib3 and friends
 ### Skills
 
 Codee uses skills to perform tasks. It is the same skills which are used in Claude Code, Codex and other agents.
-
-
-TODO: add skill types, x-codee skills extension

{codee_agent-0.5.0 → codee_agent-0.5.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "codee-agent"
-version = "0.5.0"
+version = "0.5.2"
 description = "A virtual co-worker that picks up Jira and Azure DevOps tasks and solves them with coding agents."
 readme = "README.md"
 license = "MIT"

{codee_agent-0.5.0 → codee_agent-0.5.2}/src/codee/admin.py

@@ -2215,7 +2215,9 @@ def task_filter_setting() -> rx.Component:
                          rows="2", width="100%")),
         hint=rx.text("Optional. Added to every task query as one more AND "
                      "condition, on top of the work items above. Leave it "
-                     "empty to pick up everything they match.",
+                     "empty to pick up everything they match. The poll does "
+                     "not filter on an assignee — put one here if you want "
+                     "one.",
                      color=MUTED, font_size="0.8rem"))
 
 

{codee_agent-0.5.0 → codee_agent-0.5.2}/src/codee/admin_service.py

@@ -380,6 +380,15 @@ def _remove_redundant_skill_transitions(
         for transition in transitions:
             if transition is candidate or transition["label"].casefold() != label:
                 continue
+            # Another entry for the same pair is a duplicate, not a detour.
+            # Counting it would make each copy the "longer path" that deletes
+            # the other, and the edge would vanish entirely. A skill that
+            # states one transition in several places (story-planner names
+            # `AI Decomposition review` for both a finished plan and an open
+            # question) is exactly what makes the agent emit those copies.
+            if (transition["source"].casefold() == source
+                    and transition["target"].casefold() == target):
+                continue
             adjacency.setdefault(transition["source"].casefold(), set()).add(
                 transition["target"].casefold()
             )

{codee_agent-0.5.0 → codee_agent-0.5.2}/src/codee/templates/skills/story-code-reviewer/SKILL.md

@@ -3,7 +3,7 @@ name: story-code-reviewer
 description: Review a pull request for one subtask in an Issue Tracker story.
 disable-model-invocation: true
 x-codee-trigger: issue
-x-codee-issue-status: ['AI CR Needed']
+x-codee-issue-status: ['AI Ready for CR']
 x-codee-issue-type: story
 argument-hint: <STORY_ID>
 ---
@@ -19,6 +19,13 @@ argument-hint: <STORY_ID>
 5. Review security, correctness, requirement coverage, performance, tests, error handling, and maintainability.
 6. Post line comments for specific defects and a findings-first pull request review.
 7. Add a concise Issue Tracker comment with the verdict, blocking findings, and verification performed.
+8. Move the story to `AI Ready for security review` when the review passes, or back to `AI Ready for development` when it does not.
+
+## Status Transitions
+
+- Review passed with no blocking findings: move the story to `AI Ready for security review`.
+- Blocking findings, or a relevant failing pipeline: move the story to `AI Ready for development` so the developer skill addresses them.
+- Subtasks still awaiting review: leave the story in `AI Ready for CR` so the next run reviews the next one.
 
 ## Guidelines
 
@@ -26,4 +33,4 @@ argument-hint: <STORY_ID>
 - Distinguish blocking defects from optional suggestions.
 - Do not approve changes with unresolved blocking findings.
 - Do not approve a failing pipeline unless the failure is clearly unrelated.
-- Make feedback actionable by naming the observed behavior, expected behavior, and affected location.
+- Make feedback actionable by naming the observed behavior, expected behavior, and affected location.

codee_agent-0.5.2/src/codee/templates/skills/story-developer/SKILL.md

@@ -0,0 +1,34 @@
+---
+name: story-developer
+description: Implement one subtask from an Issue Tracker story and submit a pull request.
+disable-model-invocation: true
+x-codee-trigger: issue
+x-codee-issue-status: ['AI Ready for development', 'AI In Progress']
+x-codee-issue-type: story
+argument-hint: <STORY_ID>
+---
+
+# Story Developer
+
+## Workflow
+
+1. Read the story, acceptance criteria, attachments, comments, and subtasks in the Issue Tracker.
+2. Read `story-spec/{STORY_ID}/README.md` and the affected projects' `CLAUDE.md`, `AGENTS.md`, and relevant local skills.
+3. Move the story to `AI In Progress` before doing any work, so a story already being worked on is not picked up as new.
+4. Select one open subtask and inspect any related branches, pull requests, and review comments.
+5. For bugs and regressions, reproduce the problem and identify the root cause before editing code.
+6. Create or reuse a dedicated worktree and branch according to repository conventions.
+7. Install dependencies, implement the smallest complete change, and update the story specification when useful.
+8. Run focused tests, the project build, and any required browser checks.
+9. Review the complete diff for correctness, security, regressions, and missing tests.
+10. Commit, push, and create or update the pull request. Do not merge it unless explicitly requested.
+11. Add a concise Issue Tracker comment describing the change, validation, pull request, and any remaining risk.
+12. Move the story to `AI Ready for CR` once every open subtask is implemented and its pull request is open.
+
+## Status Transitions
+
+- Work started: move the story to `AI In Progress`.
+- All subtasks implemented and pushed: move the story to `AI Ready for CR`.
+- Open subtasks remain: leave the story in `AI In Progress` so the next run continues the work.
+
+For UI work, include screenshots or video evidence when appropriate.

{codee_agent-0.5.0 → codee_agent-0.5.2}/src/codee/templates/skills/story-planner/SKILL.md

@@ -3,7 +3,7 @@ name: story-planner
 description: Decompose an Issue Tracker story into actionable subtasks and supporting documentation.
 disable-model-invocation: true
 x-codee-trigger: issue
-x-codee-issue-status: ['AI Decomposition Needed']
+x-codee-issue-status: ['AI Decomposition needed']
 x-codee-issue-type: story
 argument-hint: <STORY_ID>
 ---
@@ -29,8 +29,15 @@ Decompose an Issue Tracker story into actionable subtasks with supporting docume
 
 When a plan already exists, re-read new comments and edit the existing subtasks and specification. Create new subtasks only for newly identified work, and avoid duplicates. Move the story back to `AI Decomposition review` when the round is done.
 
+## Status Transitions
+
+- Plan posted: move the story to `AI Decomposition review`.
+- Waiting on answers to clarifying questions: move the story to `AI Decomposition review` and stop.
+
+Never move the story to `AI Ready for development` yourself. Approving the plan is a human decision.
+
 ## Guidelines
 
 - Prefer subtasks that can be implemented and reviewed independently.
 - Order subtasks by dependency.
-- Keep technical details concrete and acceptance criteria verifiable.
+- Keep technical details concrete and acceptance criteria verifiable.

{codee_agent-0.5.0 → codee_agent-0.5.2}/src/codee/templates/skills/story-qa/SKILL.md

@@ -24,5 +24,12 @@ Validate one subtask per invocation. QA is read-only: do not modify the implemen
 8. Mark each criterion pass or fail and separate blocking defects from non-blocking observations.
 9. Post a concise Issue Tracker report with the results, evidence, environment, and actionable reproduction steps for failures.
 10. Update the story specification with a short QA summary when useful.
+11. Move the story to `AI Ready for Human review` when every criterion passes, or back to `AI Ready for development` when any blocking defect was found.
 
-Create a separate Issue Tracker item for unrelated defects instead of expanding the subtask.
+## Status Transitions
+
+- Every applicable acceptance criterion passed: move the story to `AI Ready for Human review`.
+- Any blocking defect: move the story to `AI Ready for development` with the reproduction steps.
+- Subtasks still awaiting verification: leave the story in `AI Ready for QA` so the next run validates the next one.
+
+Create a separate Issue Tracker item for unrelated defects instead of expanding the subtask.

codee_agent-0.5.2/src/codee/templates/skills/story-security-reviewer/SKILL.md

@@ -0,0 +1,38 @@
+---
+name: story-security-reviewer
+description: Run a security review of the pull request for one subtask in an Issue Tracker story.
+disable-model-invocation: true
+x-codee-trigger: issue
+x-codee-issue-status: ['AI Ready for security review']
+x-codee-issue-type: story
+argument-hint: <STORY_ID>
+---
+
+# Story Security Reviewer
+
+A security-only pass over work that already passed code review. In a general review, security findings compete with naming, tests, and performance and usually lose. This run looks at nothing else.
+
+## Workflow
+
+1. Read the story, acceptance criteria, attachments, comments, and linked subtasks in the Issue Tracker.
+2. Read `story-spec/{STORY_ID}/README.md` when it exists.
+3. Select one subtask whose pull request has passed code review.
+4. Read the project instructions, complete diff, affected files, and the trust boundaries the change touches.
+5. Review authentication, authorization, input validation, injection, secrets and credential handling, sensitive data exposure, unsafe deserialization, path and command handling, dependency risk, and access to third-party services.
+6. For each finding, record the vulnerable location, the attack it enables, and the concrete fix.
+7. Post line comments for specific vulnerabilities and a findings-first pull request review.
+8. Add a concise Issue Tracker comment with the verdict, blocking findings, and what was inspected.
+9. Move the story to `AI Ready for QA` when the review passes, or back to `AI Ready for development` when it does not.
+
+## Status Transitions
+
+- No exploitable findings: move the story to `AI Ready for QA`.
+- Any exploitable finding: move the story to `AI Ready for development` with the location and the fix.
+- Subtasks still awaiting a security pass: leave the story in `AI Ready for security review` so the next run reviews the next one.
+
+## Guidelines
+
+- Review one subtask per invocation.
+- Report only findings reachable in this codebase, and name the path that reaches them.
+- Separate exploitable vulnerabilities from hardening suggestions.
+- Do not modify the implementation or merge the pull request.

codee_agent-0.5.2/src/codee/templates/skills/task-code-reviewer/SKILL.md

@@ -0,0 +1,33 @@
+---
+name: task-code-reviewer
+description: Review the pull request for a standalone Issue Tracker task.
+disable-model-invocation: true
+x-codee-trigger: issue
+x-codee-issue-status: ['AI Ready for CR']
+x-codee-issue-type: task
+argument-hint: <TASK_ID>
+---
+
+# Task Code Reviewer
+
+## Workflow
+
+1. Read the task, acceptance criteria, attachments, and comments in the Issue Tracker.
+2. Locate the pull request awaiting review.
+3. Read the project instructions, complete diff, affected files, relevant callers, tests, pipeline results, and unresolved review comments.
+4. Review security, correctness, requirement coverage, performance, tests, error handling, and maintainability.
+5. Post line comments for specific defects and a findings-first pull request review.
+6. Add a concise Issue Tracker comment with the verdict, blocking findings, and verification performed.
+7. Move the task to `AI Ready for security review` when the review passes, or back to `AI Ready for development` when it does not.
+
+## Status Transitions
+
+- Review passed with no blocking findings: move the task to `AI Ready for security review`.
+- Blocking findings, or a relevant failing pipeline: move the task to `AI Ready for development` so the developer skill addresses them.
+
+## Guidelines
+
+- Distinguish blocking defects from optional suggestions.
+- Do not approve changes with unresolved blocking findings.
+- Do not approve a failing pipeline unless the failure is clearly unrelated.
+- Make feedback actionable by naming the observed behavior, expected behavior, and affected location.

codee_agent-0.5.2/src/codee/templates/skills/task-developer/SKILL.md

@@ -0,0 +1,33 @@
+---
+name: task-developer
+description: Implement a standalone Issue Tracker task and submit a pull request.
+disable-model-invocation: true
+x-codee-trigger: issue
+x-codee-issue-status: ['AI Ready for development', 'AI In Progress']
+x-codee-issue-type: task
+argument-hint: <TASK_ID> [continue]
+---
+
+# Task Developer
+
+## Workflow
+
+1. Read the task, acceptance criteria, attachments, and comments in the Issue Tracker. With `continue`, also read new pull request feedback.
+2. Read the affected projects' `CLAUDE.md`, `AGENTS.md`, and relevant local skills.
+3. Move the task to `AI In Progress` before doing any work, so a task already being worked on is not picked up as new.
+4. For bugs and regressions, reproduce the problem and identify the root cause before editing code. Report evidence when the issue cannot be reproduced or the root cause remains unclear.
+5. Create or reuse a dedicated worktree and branch according to repository conventions.
+6. Install dependencies and implement the smallest complete change that satisfies the acceptance criteria.
+7. Run focused tests, the project build, and browser validation for UI work.
+8. Review the complete diff for correctness, security, regressions, and missing tests.
+9. Commit, push, and create or update the pull request. Do not merge it unless explicitly requested.
+10. Add a concise Issue Tracker comment describing the change, validation, pull request, and any remaining risk.
+11. Move the task to `AI Ready for CR` once the change is complete and its pull request is open.
+
+## Status Transitions
+
+- Work started: move the task to `AI In Progress`.
+- Change complete and pushed: move the task to `AI Ready for CR`.
+- Work unfinished, or the problem could not be reproduced: leave the task in `AI In Progress` and comment with what is missing.
+
+For UI work, include screenshots or video evidence when appropriate. On continued work, always refresh the Issue Tracker item and pull request before acting.

{codee_agent-0.5.0 → codee_agent-0.5.2}/src/codee/templates/skills/task-qa/SKILL.md

@@ -22,5 +22,11 @@ QA is read-only: do not modify the implementation or merge its pull request.
 6. For service work, test success, validation, authorization, and edge cases. Record requests, responses, HTTP codes, and relevant side effects.
 7. Mark each criterion pass or fail and separate blocking defects from non-blocking observations.
 8. Post a concise Issue Tracker report with the results, evidence, environment, and actionable reproduction steps for failures.
+9. Move the task to `AI Ready for Human review` when every criterion passes, or back to `AI Ready for development` when any blocking defect was found.
 
-Create a separate Issue Tracker item for unrelated defects instead of expanding the task.
+## Status Transitions
+
+- Every acceptance criterion passed: move the task to `AI Ready for Human review`.
+- Any blocking defect: move the task to `AI Ready for development` with the reproduction steps.
+
+Create a separate Issue Tracker item for unrelated defects instead of expanding the task.

codee_agent-0.5.2/src/codee/templates/skills/task-security-reviewer/SKILL.md

@@ -0,0 +1,35 @@
+---
+name: task-security-reviewer
+description: Run a security review of the pull request for a standalone Issue Tracker task.
+disable-model-invocation: true
+x-codee-trigger: issue
+x-codee-issue-status: ['AI Ready for security review']
+x-codee-issue-type: task
+argument-hint: <TASK_ID>
+---
+
+# Task Security Reviewer
+
+A security-only pass over work that already passed code review. In a general review, security findings compete with naming, tests, and performance and usually lose. This run looks at nothing else.
+
+## Workflow
+
+1. Read the task, acceptance criteria, attachments, and comments in the Issue Tracker.
+2. Locate the pull request that has passed code review.
+3. Read the project instructions, complete diff, affected files, and the trust boundaries the change touches.
+4. Review authentication, authorization, input validation, injection, secrets and credential handling, sensitive data exposure, unsafe deserialization, path and command handling, dependency risk, and access to third-party services.
+5. For each finding, record the vulnerable location, the attack it enables, and the concrete fix.
+6. Post line comments for specific vulnerabilities and a findings-first pull request review.
+7. Add a concise Issue Tracker comment with the verdict, blocking findings, and what was inspected.
+8. Move the task to `AI Ready for QA` when the review passes, or back to `AI Ready for development` when it does not.
+
+## Status Transitions
+
+- No exploitable findings: move the task to `AI Ready for QA`.
+- Any exploitable finding: move the task to `AI Ready for development` with the location and the fix.
+
+## Guidelines
+
+- Report only findings reachable in this codebase, and name the path that reaches them.
+- Separate exploitable vulnerabilities from hardening suggestions.
+- Do not modify the implementation or merge the pull request.

{codee_agent-0.5.0 → codee_agent-0.5.2}/src/codee/test_admin_service.py

@@ -13,8 +13,8 @@ from codee_agent_claude_code.provider import ClaudeCodeAgent
 from codee_agent_github_copilot.provider import GitHubCopilotAgent
 
 from codee.admin_service import (
-    MCP_CHECK, TASKS_CHECK, AdminService, azure_oauth, normalize_work_items,
-    parse_skill, repository_name)
+    MCP_CHECK, TASKS_CHECK, AdminService, _remove_redundant_skill_transitions,
+    azure_oauth, normalize_work_items, parse_skill, repository_name)
 from codee_main_context.context import (
     CodeeMainContext, CodingAgent, Settings, TasksProvider, load_settings,
     save_settings)
@@ -1676,5 +1676,70 @@ class SetupTasksMcpTest(unittest.TestCase):
             self.assertIn("not valid JSON", message)
 
 
+class RemoveRedundantSkillTransitionsTest(unittest.TestCase):
+    """Only a genuine detour removes an edge, never a repeated statement."""
+
+    def _transition(self, source: str, target: str, label: str,
+                    evidence: str) -> dict[str, str]:
+        return {"source": source, "target": target,
+                "label": label, "evidence": evidence}
+
+    def test_a_bypass_of_a_two_step_path_is_removed(self) -> None:
+        transitions = [
+            self._transition("Ready", "In progress", "dev", "first"),
+            self._transition("In progress", "Review", "dev", "second"),
+            self._transition("Ready", "Review", "dev", "third"),
+        ]
+
+        retained = _remove_redundant_skill_transitions(transitions)
+
+        self.assertEqual([(row["source"], row["target"]) for row in retained],
+                         [("Ready", "In progress"), ("In progress", "Review")])
+
+    def test_repeating_one_transition_keeps_it(self) -> None:
+        """A skill that names the same target twice still gets its edge.
+
+        story-planner moves the story to `AI Decomposition review` both when
+        the plan is ready and when it stops to ask a question, so the agent
+        reports the pair twice. Reading the second copy as a longer path made
+        each delete the other and the transition disappeared from the graph.
+        """
+        transitions = [
+            self._transition("AI Decomposition needed", "AI Decomposition review",
+                             "story-planner", "plan posted"),
+            self._transition("AI Decomposition needed", "AI Decomposition review",
+                             "story-planner", "questions asked"),
+        ]
+
+        retained = _remove_redundant_skill_transitions(transitions)
+
+        self.assertEqual(retained, transitions)
+
+    def test_a_duplicated_bypass_is_still_removed(self) -> None:
+        transitions = [
+            self._transition("Ready", "In progress", "dev", "first"),
+            self._transition("In progress", "Review", "dev", "second"),
+            self._transition("Ready", "Review", "dev", "third"),
+            self._transition("Ready", "Review", "dev", "fourth"),
+        ]
+
+        retained = _remove_redundant_skill_transitions(transitions)
+
+        self.assertEqual([(row["source"], row["target"]) for row in retained],
+                         [("Ready", "In progress"), ("In progress", "Review")])
+
+    def test_another_skill_bypass_is_left_alone(self) -> None:
+        """The detour has to belong to the same skill to count as one."""
+        transitions = [
+            self._transition("Ready", "In progress", "dev", "first"),
+            self._transition("In progress", "Review", "dev", "second"),
+            self._transition("Ready", "Review", "qa", "third"),
+        ]
+
+        retained = _remove_redundant_skill_transitions(transitions)
+
+        self.assertEqual(len(retained), 3)
+
+
 if __name__ == "__main__":
     unittest.main()

{codee_agent-0.5.0 → codee_agent-0.5.2}/src/codee_tasks_abstract/provider.py

@@ -146,7 +146,7 @@ class AbstractTasksProvider(ABC):
         """
         if not tasks:
             return (f"Connected to {self.describe()}. "
-                    "No tasks are assigned to it right now.")
+                    "No tasks are waiting for it right now.")
         preview = ", ".join(f"{task.key} {task.summary}" for task in tasks[:3])
         more = f", +{len(tasks) - 3} more" if len(tasks) > 3 else ""
         return (f"Connected to {self.describe()}. "

{codee_agent-0.5.0 → codee_agent-0.5.2}/src/codee_tasks_azure_devops/provider.py

@@ -1,7 +1,7 @@
 """Read-only Azure DevOps tasks provider, authenticated through Entra ID.
 
-Every call here is a read: a WIQL query for the ids assigned to the connected
-account, then a batch fetch of those work items. The WIQL endpoint is a POST,
+Every call here is a read: a WIQL query for the ids of the Codee work items in
+the states the skills trigger on, then a batch fetch of those work items. The WIQL endpoint is a POST,
 but it is a query — nothing in this module creates or modifies a work item.
 """
 from urllib.parse import quote
@@ -36,8 +36,8 @@ _FIELDS = [
 # Which Azure DevOps work item types Codee picks up, and what each stands for,
 # is configured per installation in Settings — organizations disagree on the
 # names ("User Story" in Agile, "Product Backlog Item" in Scrum, anything at
-# all in a custom process). Anything else assigned to the connected account
-# belongs to a human and is left alone.
+# all in a custom process). A work item of any other type belongs to a human
+# and is left alone, whatever state it reaches.
 
 # Azure DevOps priority is 1-4 with 1 highest; the executor logs this next to
 # JIRA-style names, so translate rather than print a bare digit.
@@ -122,7 +122,7 @@ class AzureDevOpsWorkItem(Task):
 
 
 class AzureDevOpsTasksProvider(AbstractTasksProvider):
-    """Fetches work items assigned to the connected account as provider-agnostic Tasks."""
+    """Fetches the organization's Codee work items as provider-agnostic Tasks."""
 
     DISPLAY_NAME = "Azure DevOps"
     MCP_SERVER_NAME = "ado"
@@ -157,11 +157,14 @@ class AzureDevOpsTasksProvider(AbstractTasksProvider):
         connection = self._auth.connection() or {}
         account = connection.get("account") or "connected account"
         types = ", ".join(self._work_item_types.values()) or "no work item types"
+        # The account is named as the identity the query runs as, not as a
+        # filter — the poll matches on type and state, whoever a work item is
+        # assigned to.
         # The filter only gets a mention when there is one: it is off for most
         # installations, and "filter none" reads like a setting gone wrong.
         extra = f", filter {self._task_filter}" if self._task_filter else ""
         return (f"Azure DevOps {self._config.organization_url} "
-                f"(all projects, assignee {account}, types {types}{extra})")
+                f"(all projects, connected as {account}, types {types}{extra})")
 
     def mcp_server(self) -> McpServer | None:
         """Microsoft's Azure DevOps MCP server, addressed at this organization.
@@ -204,7 +207,7 @@ class AzureDevOpsTasksProvider(AbstractTasksProvider):
         return [
             f'Create a new "{item_type}" work item in the {organization} '
             "organization, in any project you can create work items in, with "
-            f'the title "{summary}", assigned to {account}.',
+            f'the title "{summary}".',
             "Move that work item to a Done, Closed or Removed state — "
             "whichever its board offers — so it does not stay open.",
         ]
@@ -280,7 +283,7 @@ class AzureDevOpsTasksProvider(AbstractTasksProvider):
 
     def get_tasks(self, statuses: list[str],
                   raise_errors: bool = False) -> list[Task]:
-        """Fetch work items assigned to the connected account in the given states."""
+        """Fetch the Codee work items sitting in the given states."""
         # Nothing is waiting on a work item, so there is no request worth
         # making. The settings check passes no statuses too, but there the whole
         # point is to reach Azure DevOps, so it queries without a state filter.
@@ -342,29 +345,39 @@ class AzureDevOpsTasksProvider(AbstractTasksProvider):
         """The state filter, dropped entirely when nothing was requested.
 
         ``IN ()`` is not valid WIQL, and the only caller that passes no statuses
-        is the connection check — it wants every Codee work item assigned to the
-        account, whatever state it sits in.
+        is the connection check — it wants every Codee work item it can see,
+        whatever state it sits in.
         """
         if not statuses:
             return ""
         quoted = ", ".join(_quote_wiql(status) for status in statuses)
-        return f"AND [System.State] IN ({quoted}) "
+        return f"[System.State] IN ({quoted})"
 
     def _build_wiql(self, statuses: list[str]) -> str:
-        """WIQL for Codee work items owned by the connected account, highest priority first.
-
-        Nothing here names a project, and no ``@project`` macro is used — that
-        is what keeps the query valid with no project in the route, so it spans
-        the organization. What comes back is still narrow: only the Codee work
-        item types, only the states asked for, and only what is assigned to the
-        connected account.
+        """WIQL for Codee work items, highest priority first, then oldest.
+
+        There is no assignee clause: what hands a work item to Codee is its
+        type and its state, not who it is assigned to. So a work item a human
+        still owns is picked up the moment it reaches a state one of the skills
+        triggers on, which is what makes those states the handover — they have
+        to be ones only Codee's workflow uses. An installation that does want
+        an owner filter writes one as the custom WIQL in Settings, e.g.
+        ``[System.AssignedTo] = @Me``.
+
+        Nothing here names a project either, and no ``@project`` macro is used
+        — that is what keeps the query valid with no project in the route, so
+        it spans the organization. What comes back is still narrow: only the
+        Codee work item types, and only the states asked for.
         """
+        clauses = [clause for clause in (
+            self._build_wiql_type_clause(),
+            self._build_wiql_status_clause(statuses),
+            self._build_wiql_filter_clause(),
+        ) if clause]
+        where = f"WHERE {' AND '.join(clauses)} " if clauses else ""
         return (
             "SELECT [System.Id] FROM WorkItems "
-            "WHERE [System.AssignedTo] = @Me "
-            f"{self._build_wiql_type_clause()}"
-            f"{self._build_wiql_status_clause(statuses)}"
-            f"{self._build_wiql_filter_clause()}"
+            f"{where}"
             "ORDER BY [Microsoft.VSTS.Common.Priority] ASC, [System.CreatedDate] ASC"
         )
 
@@ -373,25 +386,26 @@ class AzureDevOpsTasksProvider(AbstractTasksProvider):
 
         Bracketed, because a filter is a whole condition rather than a single
         term: an unparenthesized ``... OR ...`` would bind its OR across the
-        assignee and type clauses and hand back items Codee does not own.
+        type and state clauses and hand back items Codee does not own.
         """
         if not self._task_filter:
             return ""
-        return f"AND ({self._task_filter}) "
+        return f"({self._task_filter})"
 
     def _build_wiql_type_clause(self) -> str:
         """The work item type filter, from the work items configured in Settings.
 
         Dropped when nothing is mapped, like the state clause above and for the
-        same reason — but this one dropping is worse than a wide query: it would
-        hand the executor every item assigned to the account. Settings keeps the
-        mandatory work items filled in so it cannot happen in practice.
+        same reason — but this one dropping is worse than a wide query: with no
+        assignee clause to fall back on it would hand the executor every item in
+        the organization. Settings keeps the mandatory work items filled in so
+        it cannot happen in practice.
         """
         item_types = list(self._work_item_types.values())
         if not item_types:
             return ""
         quoted = ", ".join(_quote_wiql(item_type) for item_type in item_types)
-        return f"AND [System.WorkItemType] IN ({quoted}) "
+        return f"[System.WorkItemType] IN ({quoted})"
 
     def _fetch_work_items(self, token: str, ids: list[int]) -> list[dict]:
         """Batch-fetch the requested work items. Organization-scoped, as the API requires."""

{codee_agent-0.5.0 → codee_agent-0.5.2}/src/codee_tasks_azure_devops/test.py

@@ -339,9 +339,16 @@ class TasksProviderTest(unittest.TestCase):
         wiql = self.provider._build_wiql(["Ready", "Bob's queue"])
 
         self.assertIn("[System.State] IN ('Ready', 'Bob''s queue')", wiql)
-        self.assertIn("[System.AssignedTo] = @Me", wiql)
         self.assertIn("ORDER BY [Microsoft.VSTS.Common.Priority] ASC", wiql)
 
+    def test_the_query_does_not_filter_on_an_assignee(self) -> None:
+        # Type and state are the handover; who the work item is assigned
+        # to is nobody's business but the humans working alongside it.
+        wiql = self.provider._build_wiql(["Ready"])
+
+        self.assertNotIn("System.AssignedTo", wiql)
+        self.assertNotIn("@Me", wiql)
+
     def test_wiql_only_asks_for_the_mapped_work_item_types(self) -> None:
         wiql = self.provider._build_wiql(["Ready"])
 
@@ -534,7 +541,7 @@ class TasksProviderTest(unittest.TestCase):
         wiql = self.provider._build_wiql([])
 
         self.assertNotIn("[System.State] IN", wiql)
-        self.assertIn("[System.AssignedTo] = @Me", wiql)
+        self.assertIn("[System.WorkItemType] IN", wiql)
 
     def test_a_successful_pull_names_the_work_items_it_found(self) -> None:
         self._connect()
@@ -614,7 +621,8 @@ class AzureDevOpsMcpTest(unittest.TestCase):
         self.assertIn('"Task" work item', steps[0])
         self.assertIn("acme organization", steps[0])
         self.assertIn('title "Codee check 1234"', steps[0])
-        self.assertIn("assigned to dev@acme.com", steps[0])
+        # Nothing is said about the assignee: the poll does not filter on one.
+        self.assertNotIn("assigned to", steps[0])
         self.assertIn("Done, Closed or Removed", steps[1])
 
     def test_no_check_steps_before_the_account_is_known(self) -> None:

{codee_agent-0.5.0 → codee_agent-0.5.2}/src/codee_tasks_jira/provider.py

@@ -174,7 +174,7 @@ class JiraTasksProvider(AbstractTasksProvider):
 
     def get_tasks(self, statuses: list[str],
                   raise_errors: bool = False) -> list[Task]:
-        """Fetch tasks assigned to the target user in the configured states."""
+        """Fetch the Codee issues sitting in the configured statuses."""
         # Nothing is waiting on an issue, so there is no request worth making.
         # The settings check passes no statuses too, but there the whole point
         # is to reach JIRA, so it queries without a status filter.

codee_agent-0.5.0/src/codee/templates/skills/story-developer/SKILL.md

@@ -1,26 +0,0 @@
----
-name: story-developer
-description: Implement one subtask from an Issue Tracker story and submit a pull request.
-disable-model-invocation: true
-x-codee-trigger: issue
-x-codee-issue-status: ['AI Ready for development', 'AI In Progress']
-x-codee-issue-type: story
-argument-hint: <STORY_ID>
----
-
-# Story Developer
-
-## Workflow
-
-1. Read the story, acceptance criteria, attachments, comments, and subtasks in the Issue Tracker.
-2. Read `story-spec/{STORY_ID}/README.md` and the affected projects' `CLAUDE.md`, `AGENTS.md`, and relevant local skills.
-3. Select one open subtask and inspect any related branches, pull requests, and review comments.
-4. For bugs and regressions, reproduce the problem and identify the root cause before editing code.
-5. Create or reuse a dedicated worktree and branch according to repository conventions.
-6. Install dependencies, implement the smallest complete change, and update the story specification when useful.
-7. Run focused tests, the project build, and any required browser checks.
-8. Review the complete diff for correctness, security, regressions, and missing tests.
-9. Commit, push, and create or update the pull request. Do not merge it unless explicitly requested.
-10. Add a concise Issue Tracker comment describing the change, validation, pull request, and any remaining risk.
-
-For UI work, include screenshots or video evidence when appropriate.

codee_agent-0.5.0/src/codee/templates/skills/task-developer/SKILL.md

@@ -1,25 +0,0 @@
----
-name: task-developer
-description: Implement a standalone Issue Tracker task and submit a pull request.
-disable-model-invocation: true
-x-codee-trigger: issue
-x-codee-issue-status: ['AI Ready for development', 'AI In Progress', Reopened]
-x-codee-issue-type: task
-argument-hint: <TASK_ID> [continue]
----
-
-# Task Developer
-
-## Workflow
-
-1. Read the task, acceptance criteria, attachments, and comments in the Issue Tracker. With `continue`, also read new pull request feedback.
-2. Read the affected projects' `CLAUDE.md`, `AGENTS.md`, and relevant local skills.
-3. For bugs and regressions, reproduce the problem and identify the root cause before editing code. Report evidence when the issue cannot be reproduced or the root cause remains unclear.
-4. Create or reuse a dedicated worktree and branch according to repository conventions.
-5. Install dependencies and implement the smallest complete change that satisfies the acceptance criteria.
-6. Run focused tests, the project build, and browser validation for UI work.
-7. Review the complete diff for correctness, security, regressions, and missing tests.
-8. Commit, push, and create or update the pull request. Do not merge it unless explicitly requested.
-9. Add a concise Issue Tracker comment describing the change, validation, pull request, and any remaining risk.
-
-For UI work, include screenshots or video evidence when appropriate. On continued work, always refresh the Issue Tracker item and pull request before acting.