Marshall

Test agent connectors end-to-end, including auth, read access, safe write/send checks, and whether a connector is visible inside an agent session. Use when Codex needs to verify launch-readiness for connectors such as Google Workspace, Telegram, email-style connectors, or similar integrations, and label the result as working, partial, or broken.

Run a practical pre-launch review for agent systems and decide whether the system is ready for real users. Use when Codex needs to evaluate runtime reliability, connector readiness, approval policies, fallback behavior, observability, safety checks, rollout readiness, operator handoff, known failure modes, and produce a go / no-go launch decision.

Design practical approval policies for agent actions, including what should always ask, what can auto-run, and what should never run. Use when Codex needs to help a team or builder define trust boundaries for local execution, messaging, email, file edits, external actions, or rollout modes such as strict, balanced, and experimental.

Onboard non-technical business teams to Codex with a controlled rollout, approval model, trust boundaries, role-based workflows, and operational examples. Use when Codex needs to help a company introduce AI to accounting, operations, admin, wholesale-market, or other back-office teams; design first-week rollout plans; define what AI may draft vs what humans must approve; convert recurring work into safe prompts and operating procedures; or produce practical onboarding materials for managers and staff.

Help finance and accounting teams use Codex for safe draft-first workflows such as request triage, document and statement summaries, checklist-based close support, internal note drafting, and approval-aware task preparation. Use when Codex needs to support non-technical or lightly technical accounting users without pretending to replace an accountant, post entries, or provide legal or tax advice.

Help operations teams use Codex for practical daily work such as inbox triage, summaries, follow-up drafting, task breakdown, recurring operational workflows, approvals, and handoff prep. Use when Codex needs to support mixed-technical or non-technical ops users with concrete, reviewable, low-friction workflows instead of vague AI advice.

Help support and operations teams triage incoming customer requests, classify urgency and ownership, draft replies for review, prepare escalations, and separate safe automation from human-reviewed actions. Use when Codex needs to turn messy support queues into reviewable next steps without making unsupported promises to customers.

Draft clear internal and customer-facing incident communications, timeline summaries, status updates, holding replies, and post-incident follow-ups. Use when Codex needs to support incident response communication without inventing root cause, over-promising fixes, or sending unreviewed high-risk updates.

Debug a local agent runtime when runs stay queued, workers do not claim work, heartbeats stop, completion never lands, or connector-triggered executions fail locally. Use for queue/claim issues, worker registration problems, startup failures, and misleading run-status symptoms.

Set up and harden a local companion or daemon that receives remote requests, runs approved Codex-compatible tasks, and returns results safely. Use when Codex needs to define the execution path behind remote control features such as Telegram chatops, local worker flows, or controlled remote command handling.

Convert meeting notes, transcripts, Slack threads, and rough recaps into actionable tasks, owners, deadlines, decisions, risks, and follow-up drafts. Use when Codex needs to turn messy discussion into a reviewable execution plan instead of a vague summary.

Choose the highest-leverage next step in a project or task and draft the next clean prompt, handoff, or execution request. Use when Codex needs to decide what to do next, reduce ambiguity, clean up messy context into one actionable move, prioritize competing options, or generate the next prompt for itself or another agent.

Operate remote machines and services through existing remote-access tools such as SSH, cloud CLIs, admin APIs, deployment hooks, or approved scripts. Use when Codex needs to run staged diagnostics, approval-gated remote actions, restarts, redeploys, rollback steps, and operator-style incident handling without pretending it has built-in remote control.

Set up, verify, route, and debug Telegram bot agents. Use when Codex needs to configure a Telegram bot token and webhook or polling flow, verify inbound and outbound delivery, map Telegram chats to agent sessions, or diagnose why a Telegram agent appears online but does not reply.

Help teams use Codex inside Telegram-centered workflows by triaging inbound Telegram messages, preparing reply drafts, structuring handoffs, summarizing chat activity, and enforcing approval-aware reply behavior. Use when Codex needs to operate on Telegram message flows themselves rather than debug Telegram bot plumbing.

Operate Codex-backed remote workflows through Telegram as the chat surface. Use when Codex needs to receive Telegram commands, classify intent, prepare approval-aware actions, return operator-friendly status updates, and hand work off to an existing remote execution path such as a companion daemon, SSH runner, or admin API.