Reported issues for Moira
Pod holds 16 of 45 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded entirely — a maintainer's release checklist is not a warning to a prospective user.
Back to Moira.
Most discussed
fix(github): preserve repository coordinates while draining claim commands
Reproduction
- Create a claimable issue.
- Comment exactly
/claim. - Observe Issue claims run 32863757200.
Observed
The command job fails before changing issue state:
TypeError: Cannot read properties of undefined (reading 'owner')
at .github/scripts/issue-claim.cjs:1353.
Suspected class
drainIssueCommands creates a synthetic context with object spread. GitHub Script exposes context.repo as context behavi
Read the thread · 2026-08-25 · closed · 4 comments
Нативный fan-out/fan-in: параллельный запуск workflow с join-узлом
Проблема
Moira уже позволяет связать отдельные запуски через parentExecutionId, а subgraph подходит для синхронного вызова одного дочернего workflow. Но для параллельного fan-out/fan-in родительскому workflow сейчас нужен внешний координатор: он должен запустить несколько дочерних workflow, сохранить их processId, опрашивать состояния и вручную продолжить родителя через step().
Если клиентская Codex/MCP-сессия завершается, такой координатор исчезает, а родитель может навсегда остать
Read the thread · 2026-08-24 · open · external user · 3 comments
docs: align MCP tools reference with registered API
Problem
The public EN/RU MCP tools reference must describe the effective schemas registered for MCP clients, not the wider internal workflow-management handler schema. The current repair also reconciles active security-automation wording.
Expected result
- EN and RU references list all and only registered public actions and inputs.
- Examples use observable handler response envelopes and representative failure behavior.
- Internal-only actions and fields are not presented as client-call
Read the thread · 2026-08-25 · closed · 2 comments
Role-gate и передача workflow между участниками и их AI-сессиями
Проблема
Workflow может дойти до шага, для которого нужна другая компетенция и другой ответственный: например, после разработки требуется тестирование. Простая проверка доступа к workflow недостаточна — выполнение должно остановиться, найти или дождаться участника с требуемой ролью и продолжиться уже в его AI-сессии.
Часть процесса до этой границы может быть открытой: например, создать задачи в Tracker, сформировать постановку и собрать общие материалы. Но дальнейшие инструкции, контекст и
Read the thread · 2026-08-24 · open · external user · 2 comments
feat: make standard workflow progress and nightly diagnostics trustworthy
Problem
The standard execution workflows need native truthful progress and must keep consuming the active plan after revision or replan. The common agent prompt also needs a minimal pointer to the existing completion-reminder API. Separately, nightly E2E lost useful diagnostics and contained stale or policy-invalid fixtures that produced repeated failures and retries.
Scope
- add native progress to Quick Task, Robust Task, and Todo List;
- bind SDF current-unit roles to the active plan
Read the thread · 2026-08-30 · closed · 0 comments
Make bundled workflow upgrades safe for customized instances
Problem
Bundled workflow upgrades use a safe whole-resource three-way comparison, but the recovery lifecycle is incomplete. When both an instance workflow and the incoming image changed relative to the accepted bundled baseline, operators can inspect candidates and resolve a conflict only after navigating mode-specific behavior. The broad force mode overwrites the whole catalog and is unsafe for ordinary recovery. SaaS preflight lacks a portable revision-bound decision plan for fresh cutover
Read the thread · 2026-08-28 · closed · 0 comments
feat: add content-rich execution progress
Goal
Add a generic content-rich execution progress contract that workflow authors can embed in any Moira workflow and that agents/users can inspect consistently in the Web UI, PNG export, Telegram notifications, CLI, and MCP authoring surfaces.
Required behavior
- Static display-only milestone graph driven by existing execution routing and variables.
- Full task title, goal, bounded facts, summaries, details, outcomes, current work, and next action without hover-only meaning.
- Exact ac
Read the thread · 2026-08-28 · closed · 0 comments
Security dependency and code hardening program
Track the repository security-hardening work enabled by the new Security Checks gate:
- update vulnerable runtime, authentication, documentation, router, and build dependency roots in compatibility-isolated pull requests;
- remediate actionable CodeQL findings with focused regression coverage;
- handle secret-scanning findings privately without exposing values;
- finish with zero open fixable findings or exact documented external blockers.
Implementation remains split into reviewed pull reques
Read the thread · 2026-08-24 · closed · 0 comments
Most recent
Команды, роли и изоляция workflow/артефактов с SSO-интеграцией
Проблема
В self-hosted Moira нет полноценной организационной модели для совместной работы нескольких групп пользователей. Workflow, исполнения и артефакты фактически принадлежат отдельному пользователю или доступны слишком широко, поэтому сложно:
- создать команды и отразить зоны ответственности;
- назначить участникам роли, например «разработчик», «тестировщик», «руководитель команды»;
- разделить личные, ролевые и общекомандные workflow;
- безопасно передать workflow или исполнение из раз
Read the thread · 2026-08-24 · open · external user · 0 comments
feat: approval-gate node via Telegram with polling — notification with buttons, the agent polls the user's answer
Problem
Approval gates in Moira are currently implemented via the agent-directive node, where the agent ITSELF asks the user (in chat/CLI) and substitutes their decision. This requires the user to be "at the keyboard" in the same interface where the agent works. There is no way to delegate the decision to Telegram: send the user a notification with buttons and make the flow WAIT for their answer while the agent stays "parked" at the gate.
Idea
A new node type (or a mode of an existing
Read the thread · 2026-06-24 · open · 0 comments
feat: dedicated MCP method for sending a Telegram notification (agent-initiated)
Problem
Currently, sending to Telegram is only possible through the telegram-notification node inside a workflow (TelegramNotificationHandler). The agent cannot, on its own initiative and at an arbitrary moment during its work, send the user a Telegram notification — for example, to report an intermediate result, ask a question, or warn about a problem outside of a node that was pre-baked into the flow.
Idea
Add a dedicated MCP method (analogous to the existing list/start/`s
Read the thread · 2026-06-24 · open · 0 comments
HTTP metrics: cardinality explosion from raw route label (Prometheus OOM)
Problem: HTTP metrics cause Prometheus cardinality explosion
The route label on the app's HTTP metrics contains the raw request path instead of a templated route. Every unique URL becomes a permanent Prometheus time-series. Vulnerability scanners hitting random paths (/api/.aws.7z, /.well-known/about.php, etc.) inflate the series count without bound, the vast majority of which belong to a single histogram metric.
Example route values actually stored as labels:
/.well-known/a
[Read the thread](https://github.com/moira-mcp/moira/issues/65) · 2026-06-24 · open · 0 comments
### feat: Workflow version history and rollback
## Context
The current state: a version field exists in the `workflow` table (`metadata.version`, semver), there is a `diff` action in MCP `manage`, and a CLI `set-version` command with backups in `workflow-backups/`. Notes already have versioning (`noteVersion` table).
## Open items
### 1. Workflow version history table
- Create a `workflowVersion` table, analogous to `noteVersion`
- Store a snapshot of the workflow JSON on each save
- Fields: workflowId, version, snapshot (JSON), changedBy,
[Read the thread](https://github.com/moira-mcp/moira/issues/62) · 2026-06-24 · open · 0 comments
### feat: Automatic artifact tracking in workflow executions
## Description
Automatically track files/data created by the agent during workflow execution.
## Context
The manual artifacts tool is implemented (upload/update/delete/list/stats/token). But the agent must explicitly call upload. There is no auto-detection and no binding of artifacts to execution steps.
## What's needed
- Auto-detection of files created by the agent
- Binding of artifacts to specific execution steps
- Dashboard with artifacts per execution
---
_Imported from the project ba
[Read the thread](https://github.com/moira-mcp/moira/issues/61) · 2026-06-24 · open · 0 comments
### feat: Load workflows from URL and git repositories
## Description
Add the ability to import workflows by URL or from a git repository.
## Context
Token upload/download is implemented. MCP `manage` create works. But there is no import from external sources.
## What's needed
- Import a workflow by URL (JSON file)
- Import from a git repository (path to JSON)
- Validation before import
---
_Imported from the project backlog._
[Read the thread](https://github.com/moira-mcp/moira/issues/60) · 2026-06-24 · open · 0 comments
### feat: Semantic validation of completionCondition
## Description
Add a check that the agent actually fulfilled the `completionCondition`, rather than merely claiming it did.
## Context
`inputSchema` AJV validation exists (structural). `maxRetries` is supported. The system prompt requires verification. But there is no automatic semantic verification.
## What's needed
- LLM-judge or programmatic check of `completionCondition`
- Optional per-node setting (enable/disable)
- Fallback: retry on failed validation
---
_Imported from the project b
[Read the thread](https://github.com/moira-mcp/moira/issues/59) · 2026-06-24 · open · 0 comments
The remaining reports are on [the project's issue tracker](https://github.com/moira-mcp/moira/issues).