{
  "schemaVersion": "1.0",
  "item": {
    "slug": "codex-orchestration",
    "name": "codex-orchestration",
    "source": "tencent",
    "type": "skill",
    "category": "AI 智能",
    "sourceUrl": "https://clawhub.ai/shanelindsay/codex-orchestration",
    "canonicalUrl": "https://clawhub.ai/shanelindsay/codex-orchestration",
    "targetPlatform": "OpenClaw"
  },
  "install": {
    "downloadMode": "redirect",
    "downloadUrl": "/downloads/codex-orchestration",
    "sourceDownloadUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=codex-orchestration",
    "sourcePlatform": "tencent",
    "targetPlatform": "OpenClaw",
    "installMethod": "Manual import",
    "extraction": "Extract archive",
    "prerequisites": [
      "OpenClaw"
    ],
    "packageFormat": "ZIP package",
    "includedAssets": [
      "SKILL.md"
    ],
    "primaryDoc": "SKILL.md",
    "quickSetup": [
      "Download the package from Yavira.",
      "Extract the archive and review SKILL.md first.",
      "Import or place the package into your OpenClaw setup."
    ],
    "agentAssist": {
      "summary": "Hand the extracted package to your coding agent with a concrete install brief instead of figuring it out manually.",
      "steps": [
        "Download the package from Yavira.",
        "Extract it into a folder your agent can access.",
        "Paste one of the prompts below and point your agent at the extracted folder."
      ],
      "prompts": [
        {
          "label": "New install",
          "body": "I downloaded a skill package from Yavira. Read SKILL.md from the extracted folder and install it by following the included instructions. Tell me what you changed and call out any manual steps you could not complete."
        },
        {
          "label": "Upgrade existing",
          "body": "I downloaded an updated skill package from Yavira. Read SKILL.md from the extracted folder, compare it with my current installation, and upgrade it while preserving any custom configuration unless the package docs explicitly say otherwise. Summarize what changed and any follow-up checks I should run."
        }
      ]
    },
    "sourceHealth": {
      "source": "tencent",
      "status": "healthy",
      "reason": "direct_download_ok",
      "recommendedAction": "download",
      "checkedAt": "2026-04-30T16:55:25.780Z",
      "expiresAt": "2026-05-07T16:55:25.780Z",
      "httpStatus": 200,
      "finalUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=network",
      "contentType": "application/zip",
      "probeMethod": "head",
      "details": {
        "probeUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=network",
        "contentDisposition": "attachment; filename=\"network-1.0.0.zip\"",
        "redirectLocation": null,
        "bodySnippet": null
      },
      "scope": "source",
      "summary": "Source download looks usable.",
      "detail": "Yavira can redirect you to the upstream package for this source.",
      "primaryActionLabel": "Download for OpenClaw",
      "primaryActionHref": "/downloads/codex-orchestration"
    },
    "validation": {
      "installChecklist": [
        "Use the Yavira download entry.",
        "Review SKILL.md after the package is downloaded.",
        "Confirm the extracted package contains the expected setup assets."
      ],
      "postInstallChecks": [
        "Confirm the extracted package includes the expected docs or setup files.",
        "Validate the skill or prompts are available in your target agent workspace.",
        "Capture any manual follow-up steps the agent could not complete."
      ]
    },
    "downloadPageUrl": "https://openagent3.xyz/downloads/codex-orchestration",
    "agentPageUrl": "https://openagent3.xyz/skills/codex-orchestration/agent",
    "manifestUrl": "https://openagent3.xyz/skills/codex-orchestration/agent.json",
    "briefUrl": "https://openagent3.xyz/skills/codex-orchestration/agent.md"
  },
  "agentAssist": {
    "summary": "Hand the extracted package to your coding agent with a concrete install brief instead of figuring it out manually.",
    "steps": [
      "Download the package from Yavira.",
      "Extract it into a folder your agent can access.",
      "Paste one of the prompts below and point your agent at the extracted folder."
    ],
    "prompts": [
      {
        "label": "New install",
        "body": "I downloaded a skill package from Yavira. Read SKILL.md from the extracted folder and install it by following the included instructions. Tell me what you changed and call out any manual steps you could not complete."
      },
      {
        "label": "Upgrade existing",
        "body": "I downloaded an updated skill package from Yavira. Read SKILL.md from the extracted folder, compare it with my current installation, and upgrade it while preserving any custom configuration unless the package docs explicitly say otherwise. Summarize what changed and any follow-up checks I should run."
      }
    ]
  },
  "documentation": {
    "source": "clawhub",
    "primaryDoc": "SKILL.md",
    "sections": [
      {
        "title": "Codex orchestration",
        "body": "You are the orchestrator: decide the work, delegate clearly, deliver a clean result.\nWorkers do the legwork; you own judgement.\n\nThis guide is steering, not bureaucracy. Use common sense. If something is simple, just do it."
      },
      {
        "title": "Default assumptions",
        "body": "YOLO config (no approvals); web search enabled.\nPTY execution available via exec_command and write_stdin.\nCodex already knows its tools; this guide is about coordination and decomposition."
      },
      {
        "title": "Orchestrator mode (default)",
        "body": "Split work into sensible tracks.\nUse parallel workers when it helps.\nKeep the main thread for synthesis, decisions, and final output."
      },
      {
        "title": "Worker mode (only when explicitly invoked)",
        "body": "A worker prompt begins with CONTEXT: WORKER.\n\nDo only the assigned task.\nDo not spawn other workers.\nReport back crisply with evidence."
      },
      {
        "title": "Planning with update_plan",
        "body": "Use update_plan when any of these apply:\n\nMore than 2 steps.\nParallel work would help.\nThe situation is unclear, messy, or high stakes.\n\nKeep it light:\n\n3 to 6 steps max.\nShort steps, one sentence each.\nExactly one step in_progress.\nUpdate the plan when you complete a step or change direction.\nSkip the plan entirely for trivial tasks."
      },
      {
        "title": "Parallelism: \"sub-agents\" as background codex exec sessions",
        "body": "A sub-agent is a background terminal running codex exec with a focused worker prompt.\n\nUse parallel workers for:\n\nScouting and mapping (where things are, current state)\nIndependent reviews (different lenses on the same artefact)\nWeb research (sources, definitions, comparisons)\nLong-running checks (tests, builds, analyses, data pipelines)\nDrafting alternatives (outlines, rewrites, options)\n\nAvoid parallel workers that edit the same artefact. Default rule: many readers, one writer."
      },
      {
        "title": "Background PTY terminals (exec_command + write_stdin)",
        "body": "Use PTY sessions to run work without blocking the main thread.\n\nexec_command runs a command in a PTY and returns output, or a session_id if it keeps running.\nIf you get a session_id, use write_stdin to poll output or interact with the same process.\n\nPractical habits:\n\nStart long tasks with small yield_time_ms so you do not stall.\nKeep max_output_tokens modest, then poll again.\nLabel each session mentally (or in your notes) like: W1 Scout, W2 Review, W3 Research.\nDefault to non-blocking: start the worker, capture its session_id, and move on.\nIf you end your turn before it finishes, say so explicitly and offer to resume polling later.\nIf the session exits or is lost, fall back to re-run or use a persistent runner (tmux/nohup).\nIf writing output to a file, check for the file before re-polling the session.\n\nBlocking vs non-blocking (recommend non-blocking even if you plan to poll):\n\nDefault to non-blocking; poll once or twice if you need quick feedback.\nBlocking is fine only for short, predictable tasks (<30–60s).\n\nStopping jobs:\n\nPrefer graceful shutdown when possible.\nIf needed, send Ctrl+C via write_stdin."
      },
      {
        "title": "Capturing worker output (keep context small)",
        "body": "Prefer capturing only the final worker message to avoid bloating the main context.\n\nRecommended (simple):\n\nUse --output-last-message to write the final response to a file, then read it.\nExample: codex exec --skip-git-repo-check --output-last-message /tmp/w1.txt \"CONTEXT: WORKER ...\"\nIf you are outside a git repo, add --skip-git-repo-check.\n\nAlternative (structured):\n\nUse --json and filter for the final agent message.\nExample: codex exec --json \"CONTEXT: WORKER ...\" | jq -r 'select(.type==\"item.completed\" and .item.type==\"agent_message\") | .item.text'"
      },
      {
        "title": "Orchestration patterns (general-purpose)",
        "body": "Pick a pattern, then run it. Do not over-engineer."
      },
      {
        "title": "Pattern A: Triangulated review (fan-out, read-only)",
        "body": "Use when: you want multiple perspectives on the same thing.\n\nRun 2 to 4 reviewers with different lenses, then merge.\n\nExample lenses (choose what fits):\n\nClarity/structure\nCorrectness/completeness\nRisks/failure modes\nConsistency/style\nEvidence quality\nPracticality\nAccessibility/audience fit\nIf relevant: security, performance, backward compatibility\n\nDeliverable: a single ranked list with duplicates removed and clear recommendations."
      },
      {
        "title": "Pattern B: Review -> fix (serial chain)",
        "body": "Use when: you want a clean funnel.\n\nReviewer produces an issue list ranked by impact.\nImplementer addresses the top items.\nVerifier checks the result.\n\nThis works for code, documents, and analyses."
      },
      {
        "title": "Pattern C: Scout -> act -> verify (classic)",
        "body": "Use when: lack of context is the biggest risk.\n\nScout gathers the minimum context.\nOrchestrator condenses it and chooses the approach.\nImplementer executes.\nVerifier sanity-checks."
      },
      {
        "title": "Pattern D: Split by sections (fan-out, then merge)",
        "body": "Use when: work divides cleanly (sections, modules, datasets, figures).\nEach worker owns a distinct slice; merge for consistency."
      },
      {
        "title": "Pattern E: Research -> synthesis -> next actions",
        "body": "Use when: the task is primarily web search and judgement.\nWorkers collect sources in parallel; orchestrator synthesises a decision-ready brief."
      },
      {
        "title": "Pattern F: Options sprint (generate 2 to 3 good alternatives)",
        "body": "Use when: you are choosing direction (outline, methods plan, analysis, UI).\nWorkers propose options; orchestrator selects and refines one."
      },
      {
        "title": "Context: supply what workers cannot infer",
        "body": "Most failures come from missing context, not missing formatting instructions.\n\nUse a Context Pack when:\n\nthe work touches an existing project with history,\nthe goal is subtle,\nconstraints are non-obvious,\nor preferences matter.\n\nSkip it when:\n\nthe task is a simple web lookup,\na small isolated edit,\nor a straightforward one-off."
      },
      {
        "title": "Context Pack (use as much or as little as needed)",
        "body": "Goal: what \"good\" looks like.\nNon-goals: what not to do.\nConstraints: style, scope boundaries, must keep, must not change.\nPointers: key files, folders, documents, notes, links.\nPrior decisions: why things are the way they are.\nSuccess check: how we know it is done (tests, criteria, checklist).\n\nAcademic writing note:\n\nFor manuscripts or scholarly text, use APA 7 where appropriate."
      },
      {
        "title": "Worker prompt templates (neutral)",
        "body": "Prepend the Worker preamble to every worker prompt."
      },
      {
        "title": "Worker preamble (use for all workers)",
        "body": "CONTEXT: WORKER\nROLE: You are a sub-agent run by the ORCHESTRATOR. Do only the assigned task.\nRULES: No extra scope, no other workers.\nYour final output will be provided back to the ORCHESTRATOR.\n\nMinimal worker command (example):\n\ncodex exec --skip-git-repo-check --output-last-message /tmp/w1.txt \"CONTEXT: WORKER\nROLE: You are a sub-agent run by the ORCHESTRATOR. Do only the assigned task.\nRULES: No extra scope, no other workers.\nYour final output will be provided back to the ORCHESTRATOR.\nTASK: <what to do>\nSCOPE: read-only\""
      },
      {
        "title": "Reviewer worker",
        "body": "CONTEXT: WORKER\nTASK: Review <artefact> and produce improvements.\nSCOPE: read-only\nLENS: <pick one or two lenses>\nDO:\n\nInspect the artefact and note issues and opportunities.\nPrioritise what matters most.\nOUTPUT:\nTop findings (ranked, brief)\nEvidence (where you saw it)\nRecommended fixes (concise, actionable)\nOptional: quick rewrite or outline snippet\nDO NOT:\nExpand scope\nMake edits"
      },
      {
        "title": "Research worker (web search)",
        "body": "CONTEXT: WORKER\nTASK: Find and summarise reliable information on <topic>.\nSCOPE: read-only\nDO:\n\nUse web search.\nPrefer primary sources, official docs, and high-quality references.\nOUTPUT:\n5 to 10 bullet synthesis\nKey sources (with short notes on why they matter)\nUncertainty or disagreements between sources\nDO NOT:\nSpeculate beyond evidence"
      },
      {
        "title": "Implementer worker (build, write, analyse, edit)",
        "body": "CONTEXT: WORKER\nTASK: Produce <deliverable>.\nSCOPE: may edit <specific files/sections> or \"write new artefact\"\nDO:\n\nFollow the Context Pack if provided.\nMake changes proportionate to the request.\nOUTPUT:\nWhat you changed or produced\nWhere it lives (paths, filenames)\nHow to reproduce (commands, steps) if relevant\nRisks or follow-ups (brief)\nDO NOT:\nDrift into unrelated improvements"
      },
      {
        "title": "Verifier worker",
        "body": "CONTEXT: WORKER\nTASK: Verify the deliverable meets the Goal and Success check.\nSCOPE: read-only (unless explicitly allowed)\nDO:\n\nRun checks (tests, builds, analyses, reference checks) if relevant.\nLook for obvious omissions and regressions.\nOUTPUT:\nPass/fail summary\nIssues with repro steps or concrete examples\nSuggested fixes (brief)"
      },
      {
        "title": "Orchestrator habits (fast, not fussy)",
        "body": "Skim the artefact yourself before delegating.\nAsk a quick clarification if a term or goal is ambiguous.\nUse parallel workers when it reduces time or uncertainty.\nKeep instructions short and context-rich; do not paste the whole skill into worker prompts.\nIf a worker misunderstood, do not argue. Re-run with better context.\nMerge outputs into one clear result, one recommended next step, and only the necessary detail.\n\nBoss rule:\nYou do not forward raw worker output unless it is already clean. You curate it."
      }
    ],
    "body": "Codex orchestration\n\nYou are the orchestrator: decide the work, delegate clearly, deliver a clean result. Workers do the legwork; you own judgement.\n\nThis guide is steering, not bureaucracy. Use common sense. If something is simple, just do it.\n\nDefault assumptions\nYOLO config (no approvals); web search enabled.\nPTY execution available via exec_command and write_stdin.\nCodex already knows its tools; this guide is about coordination and decomposition.\nTwo modes\nOrchestrator mode (default)\nSplit work into sensible tracks.\nUse parallel workers when it helps.\nKeep the main thread for synthesis, decisions, and final output.\nWorker mode (only when explicitly invoked)\n\nA worker prompt begins with CONTEXT: WORKER.\n\nDo only the assigned task.\nDo not spawn other workers.\nReport back crisply with evidence.\nPlanning with update_plan\n\nUse update_plan when any of these apply:\n\nMore than 2 steps.\nParallel work would help.\nThe situation is unclear, messy, or high stakes.\n\nKeep it light:\n\n3 to 6 steps max.\nShort steps, one sentence each.\nExactly one step in_progress.\nUpdate the plan when you complete a step or change direction.\nSkip the plan entirely for trivial tasks.\nParallelism: \"sub-agents\" as background codex exec sessions\n\nA sub-agent is a background terminal running codex exec with a focused worker prompt.\n\nUse parallel workers for:\n\nScouting and mapping (where things are, current state)\nIndependent reviews (different lenses on the same artefact)\nWeb research (sources, definitions, comparisons)\nLong-running checks (tests, builds, analyses, data pipelines)\nDrafting alternatives (outlines, rewrites, options)\n\nAvoid parallel workers that edit the same artefact. Default rule: many readers, one writer.\n\nBackground PTY terminals (exec_command + write_stdin)\n\nUse PTY sessions to run work without blocking the main thread.\n\nexec_command runs a command in a PTY and returns output, or a session_id if it keeps running.\nIf you get a session_id, use write_stdin to poll output or interact with the same process.\n\nPractical habits:\n\nStart long tasks with small yield_time_ms so you do not stall.\nKeep max_output_tokens modest, then poll again.\nLabel each session mentally (or in your notes) like: W1 Scout, W2 Review, W3 Research.\nDefault to non-blocking: start the worker, capture its session_id, and move on.\nIf you end your turn before it finishes, say so explicitly and offer to resume polling later.\nIf the session exits or is lost, fall back to re-run or use a persistent runner (tmux/nohup).\nIf writing output to a file, check for the file before re-polling the session.\n\nBlocking vs non-blocking (recommend non-blocking even if you plan to poll):\n\nDefault to non-blocking; poll once or twice if you need quick feedback.\nBlocking is fine only for short, predictable tasks (<30–60s).\n\nStopping jobs:\n\nPrefer graceful shutdown when possible.\nIf needed, send Ctrl+C via write_stdin.\nCapturing worker output (keep context small)\n\nPrefer capturing only the final worker message to avoid bloating the main context.\n\nRecommended (simple):\n\nUse --output-last-message to write the final response to a file, then read it.\nExample: codex exec --skip-git-repo-check --output-last-message /tmp/w1.txt \"CONTEXT: WORKER ...\"\nIf you are outside a git repo, add --skip-git-repo-check.\n\nAlternative (structured):\n\nUse --json and filter for the final agent message.\nExample: codex exec --json \"CONTEXT: WORKER ...\" | jq -r 'select(.type==\"item.completed\" and .item.type==\"agent_message\") | .item.text'\nOrchestration patterns (general-purpose)\n\nPick a pattern, then run it. Do not over-engineer.\n\nPattern A: Triangulated review (fan-out, read-only)\n\nUse when: you want multiple perspectives on the same thing.\n\nRun 2 to 4 reviewers with different lenses, then merge.\n\nExample lenses (choose what fits):\n\nClarity/structure\nCorrectness/completeness\nRisks/failure modes\nConsistency/style\nEvidence quality\nPracticality\nAccessibility/audience fit\nIf relevant: security, performance, backward compatibility\n\nDeliverable: a single ranked list with duplicates removed and clear recommendations.\n\nPattern B: Review -> fix (serial chain)\n\nUse when: you want a clean funnel.\n\nReviewer produces an issue list ranked by impact.\nImplementer addresses the top items.\nVerifier checks the result.\n\nThis works for code, documents, and analyses.\n\nPattern C: Scout -> act -> verify (classic)\n\nUse when: lack of context is the biggest risk.\n\nScout gathers the minimum context.\nOrchestrator condenses it and chooses the approach.\nImplementer executes.\nVerifier sanity-checks.\nPattern D: Split by sections (fan-out, then merge)\n\nUse when: work divides cleanly (sections, modules, datasets, figures). Each worker owns a distinct slice; merge for consistency.\n\nPattern E: Research -> synthesis -> next actions\n\nUse when: the task is primarily web search and judgement. Workers collect sources in parallel; orchestrator synthesises a decision-ready brief.\n\nPattern F: Options sprint (generate 2 to 3 good alternatives)\n\nUse when: you are choosing direction (outline, methods plan, analysis, UI). Workers propose options; orchestrator selects and refines one.\n\nContext: supply what workers cannot infer\n\nMost failures come from missing context, not missing formatting instructions.\n\nUse a Context Pack when:\n\nthe work touches an existing project with history,\nthe goal is subtle,\nconstraints are non-obvious,\nor preferences matter.\n\nSkip it when:\n\nthe task is a simple web lookup,\na small isolated edit,\nor a straightforward one-off.\nContext Pack (use as much or as little as needed)\nGoal: what \"good\" looks like.\nNon-goals: what not to do.\nConstraints: style, scope boundaries, must keep, must not change.\nPointers: key files, folders, documents, notes, links.\nPrior decisions: why things are the way they are.\nSuccess check: how we know it is done (tests, criteria, checklist).\n\nAcademic writing note:\n\nFor manuscripts or scholarly text, use APA 7 where appropriate.\nWorker prompt templates (neutral)\n\nPrepend the Worker preamble to every worker prompt.\n\nWorker preamble (use for all workers)\nCONTEXT: WORKER\nROLE: You are a sub-agent run by the ORCHESTRATOR. Do only the assigned task.\nRULES: No extra scope, no other workers.\nYour final output will be provided back to the ORCHESTRATOR.\n\n\nMinimal worker command (example):\n\ncodex exec --skip-git-repo-check --output-last-message /tmp/w1.txt \"CONTEXT: WORKER\nROLE: You are a sub-agent run by the ORCHESTRATOR. Do only the assigned task.\nRULES: No extra scope, no other workers.\nYour final output will be provided back to the ORCHESTRATOR.\nTASK: <what to do>\nSCOPE: read-only\"\n\nReviewer worker\n\nCONTEXT: WORKER\nTASK: Review <artefact> and produce improvements.\nSCOPE: read-only\nLENS: <pick one or two lenses>\nDO:\n\nInspect the artefact and note issues and opportunities.\nPrioritise what matters most. OUTPUT:\nTop findings (ranked, brief)\nEvidence (where you saw it)\nRecommended fixes (concise, actionable)\nOptional: quick rewrite or outline snippet\nDO NOT:\nExpand scope\nMake edits\nResearch worker (web search)\n\nCONTEXT: WORKER\nTASK: Find and summarise reliable information on <topic>.\nSCOPE: read-only\nDO:\n\nUse web search.\nPrefer primary sources, official docs, and high-quality references. OUTPUT:\n5 to 10 bullet synthesis\nKey sources (with short notes on why they matter)\nUncertainty or disagreements between sources\nDO NOT:\nSpeculate beyond evidence\nImplementer worker (build, write, analyse, edit)\n\nCONTEXT: WORKER\nTASK: Produce <deliverable>.\nSCOPE: may edit <specific files/sections> or \"write new artefact\"\nDO:\n\nFollow the Context Pack if provided.\nMake changes proportionate to the request. OUTPUT:\nWhat you changed or produced\nWhere it lives (paths, filenames)\nHow to reproduce (commands, steps) if relevant\nRisks or follow-ups (brief)\nDO NOT:\nDrift into unrelated improvements\nVerifier worker\n\nCONTEXT: WORKER\nTASK: Verify the deliverable meets the Goal and Success check.\nSCOPE: read-only (unless explicitly allowed)\nDO:\n\nRun checks (tests, builds, analyses, reference checks) if relevant.\nLook for obvious omissions and regressions. OUTPUT:\nPass/fail summary\nIssues with repro steps or concrete examples\nSuggested fixes (brief)\nOrchestrator habits (fast, not fussy)\nSkim the artefact yourself before delegating.\nAsk a quick clarification if a term or goal is ambiguous.\nUse parallel workers when it reduces time or uncertainty.\nKeep instructions short and context-rich; do not paste the whole skill into worker prompts.\nIf a worker misunderstood, do not argue. Re-run with better context.\nMerge outputs into one clear result, one recommended next step, and only the necessary detail.\n\nBoss rule: You do not forward raw worker output unless it is already clean. You curate it."
  },
  "trust": {
    "sourceLabel": "tencent",
    "provenanceUrl": "https://clawhub.ai/shanelindsay/codex-orchestration",
    "publisherUrl": "https://clawhub.ai/shanelindsay/codex-orchestration",
    "owner": "shanelindsay",
    "version": "2.0.0",
    "license": null,
    "verificationStatus": "Indexed source record"
  },
  "links": {
    "detailUrl": "https://openagent3.xyz/skills/codex-orchestration",
    "downloadUrl": "https://openagent3.xyz/downloads/codex-orchestration",
    "agentUrl": "https://openagent3.xyz/skills/codex-orchestration/agent",
    "manifestUrl": "https://openagent3.xyz/skills/codex-orchestration/agent.json",
    "briefUrl": "https://openagent3.xyz/skills/codex-orchestration/agent.md"
  }
}