OrgX Runtime Reporting

SkillProductivity

Use when a Codex execution should report progress, artifacts, blockers, or completion state back to OrgX during a live task.

Use OrgX Runtime Reporting in Claude, ChatGPT or Ahel Desktop

Free. Sign in, add OrgX Runtime Reporting and connect your AI. About a minute.

Also: Claude Code · Cursor · Codex

Then ask your AI: use the OrgX Runtime Reporting skill

Details

Instructions available. Your AI can read the instructions. Execution depends on the setup they require.

Add Ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

OrgX Runtime ReportingStart free

What this skill tells your AI

The instructions your AI receives, as published by hashgraph-online/awesome-codex-plugins in plugins/useorgx/orgx-codex-plugin/skills/orgx-runtime-reporting/SKILL.md and read by Ahel’s review.

Use this skill when Codex should keep OrgX updated during execution.

Reporting contract

There are two reporting paths:

  • Active path: call OrgX MCP tools during the work when you know the initiative, task, decision, blocker, or artifact context.
  • Chronicle readout: for operator reporting, call get_operator_chronicle first when available and present reportingNarrative.briefMarkdown before drilling into individual entities.
  • Stale-client fallback: if bootstrap or docs advertise get_operator_chronicle but the current AI client session has not refreshed its callable tool list, immediately call orgx_recommend or _orgx_recommend with mode: "morning_brief" and present the returned reportingNarrative.briefMarkdown. Do not ask the user to reconnect before giving the report.
  • Passive backstop: Codex runtime hooks installed by orgx-wizard hooks install record compact session events and run summary-only local Work Graph reconciliation on Stop.

Do not treat hook presence as a substitute for intentional OrgX writes. Hooks answer whether OrgX was used and can write a local report automatically; MCP calls, or an explicitly opted-in successful Work Graph post, make the work durable in OrgX while the session is still fresh.

Workflow

  1. Resolve available IDs from args, env, or the current OrgX context:
  • ORGX_INITIATIVE_ID
  • ORGX_WORKSTREAM_ID
  • ORGX_TASK_ID
  • ORGX_RUN_ID
  • ORGX_CORRELATION_ID
  1. For reporting questions, retrieve the operator chronicle:
  • Use get_operator_chronicle with period: "30d" for broad clarity when it is callable in the current client.
  • Use period: "day" or period: "week" when the user asks for yesterday or this week.
  • If get_operator_chronicle is not callable in the current client, use the existing orgx_recommend / _orgx_recommend fallback with mode: "morning_brief" and the broadest supported period. Treat the direct tool as preferred, but do not block on client schema refresh.
  • Lead with reportingNarrative.briefMarkdown, then call out gaps and the next action.
  1. Emit activity at meaningful milestones:
  • intent
  • execution
  • handoff
  • blocked
  • completed
  1. Register proof of work:
  • When you produce a file, diff, document, screenshot, or report, register it as an artifact with a concrete summary.
  1. Handle blockers structurally:
  • If judgment is required, request a decision with explicit options.
  • A decision waits on a person: report the blocker as waiting on them with the decision's review_url. orgx_decide approve/reject only returns that link; never report a decision as approved or rejected yourself.
  • If context is missing, report the exact missing dependency.
  1. Close execution cleanly:
  • When the task is complete and verified, emit completion activity and update entity state if the task ID is available.
  1. If no OrgX IDs are available:
  • Continue the work, but make the final response easy for the hook reconciler to classify: name decisions, artifacts, blockers, next actions, and verification.
  • Do not claim OrgX was updated unless an MCP tool or API call actually succeeded.
  • Do not claim delegated or spawned work finished because the spawn call succeeded: that only means it started. Check it with orgx_command_status (kind: "run" and the returned run ID) until next_poll_after_ms is null.
  1. Preserve Work Graph continuity:
  • When a Work Graph report is generated, include its work_graph_fingerprint and signup_hydration.hydration_key in summaries or artifacts that are safe to store.
  • Automatic Stop-hook reconciliation writes the latest local report to ~/.config/useorgx/wizard/hooks/reports/latest-work-graph-report.json.
  • Stop-hook posting requires ORGX_HOOK_RECONCILE_POST=true or ORGX_WIZARD_HOOK_RECONCILE_POST=true plus ORGX_API_KEY.
  • Treat the fingerprint as the durable claim key that lets OrgX hydrate pre-signup audit value into a user's future workspace.
  • Never derive the fingerprint from secrets or raw transcripts that would need to leave the local machine.

Quality bar

  • Never post empty status updates.
  • Messages must be evidence-based and specific.
  • Include OrgX IDs whenever available.
  • Use source_client=codex.
  • Preserve secrets: never emit tokens, cookies, API keys, or storage state into activity, retro, hook summaries, or final reports.

Signals

GitHub stars
1k
Forks
316
Last commit
Oct 2026
Advanced
Item type
skill
Key
orgx-runtime-reporting
Source
github.com/hashgraph-online/awesome-codex-plugins