[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7CBQ2RTYNXPZ0X1R1VWH6SW":3,"skill-same:01M7CBQ2RTYNXPZ0X1R1VWH6SW":46},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":14,"repositoryHtmlUrl":28,"repositoryHomepageUrl":29,"ownerHtmlUrl":30,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":31,"install":32,"resources":45},"01M7CBQ2RTYNXPZ0X1R1VWH6SW","task-execution",null,"Execute an assigned collaboration Subtask through its Task Service lifecycle.","ai-agentcore\u002Fagentcore-sdk-nodejs","ai-agentcore","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F325854201?v=4","01M7CBPS8YQXEEFDQXNJGDXQK2","Alibaba Cloud AgentCore","01M7CBPS9N1F7NRQVS3472E43D",1,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fai-agentcore\u002Fagentcore-sdk-nodejs\u002Fblob\u002Fmain\u002Fpackages\u002Fcollaboration\u002Fworker-skills\u002Ftask-execution\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: task-execution\ndescription: Execute an assigned collaboration Subtask through its Task Service lifecycle.\nversion: \"1\"\n---\n\n# Task execution\n\nYou are an AgentCore Worker. Task Service state is authoritative.\n\n## Reload and verify\n\n1. Copy the exact Subtask ID from the current notification or explicit user request, then pass it as\n   `subtask_id` to `agentteams_get_subtask`. Never infer an ID from a title or generate one. Treat the\n   returned object as the current assignment.\n2. Copy its exact `taskId`, then call `agentteams_get_task` for the parent Task. Do not derive the\n   parent ID from a message or local path.\n3. Verify the Subtask status, `assignedTo`, description, metadata, dependencies, and any referenced\n   specification. Inspect prior Results when resuming work or handling a revision. Use\n   `agentteams_list_task_events` only for history; current Task and Subtask objects remain\n   authoritative.\n4. When the assignee identity must be checked, call `agentteams_get_team_context` and compare\n   `assignedTo.userId` with this Worker's `member.matrixUserId`. Do not infer identity from names or\n   conversation text.\n5. Use the parent Task's `metadata.language` for acknowledgements, progress, blockers, Result\n   summaries, and revisions. If it is absent, consistently use the actionable request's language.\n\nFor every exact Subtask ID in `dependsOn`, call `agentteams_list_results` with the parent `task_id`\nand that dependency `subtask_id`. Do not execute until every dependency has a Result whose `status`\nis `accepted`. Use only the exact file references from that accepted Result when dependency output\nis required; do not infer completion from Subtask status, messages, events, or local files.\nDo not independently revalidate accepted dependency Results or parent Task facts; consume only\nwhat the current assignment needs.\n\nWhen a Subtask references a specification or input file, determine whether the reference belongs\nto the parent Task or the assigned Subtask before reading it. Use only the matching file tools; do\nnot try both spaces or rewrite the reference when ownership is ambiguous.\n\n## Follow the authoritative state\n\n- `assigned`: call `agentteams_ack_subtask` with the exact `subtask_id` exactly once before\n  substantive work or external side effects.\n- `in_progress`: continue without acknowledging again.\n- `blocked`, `pending_review`, `completed`, or `cancelled`: stop and wait for new input or an\n  authoritative state change.\n\nIf acknowledgement fails or its outcome is uncertain, re-read the Subtask once. Continue only when\nit is still assigned to you and is `in_progress`. If you have confirmed the Subtask is assigned to\nyou but information needed to execute it is missing, acknowledge it first and then block it.\nIf assignment is stale or does not match your identity, stop without writing.\n\n## Execute and report\n\n- Satisfy the Subtask description, expected outcome, dependencies, and every acceptance criterion\n  present in its metadata or parent Task. Perform every explicitly required test or live check.\n- Make one deliberate verification pass with the explicitly required checks and those necessary\n  to establish the requested artifact's usability. Once these pass, do not add exploratory tests,\n  repeated checks, hashes, or separate self-check artifacts. Explanatory or summary work does not\n  imply an execution check unless requested.\n- Keep deliverables no more detailed than requested. Do not add internal IDs, acceptance tables,\n  repeated calculations, or process narration unless the deliverable requires them.\n- Report progress only when it adds useful evidence; a heartbeat is not progress evidence.\n- If work continues for roughly 10 minutes since the latest successful acknowledgement, useful\n  progress update, or heartbeat, re-read the Subtask. Call `agentteams_heartbeat_subtask` with the\n  same exact `subtask_id` only when it remains assigned to you and `in_progress`. Do not start a\n  heartbeat daemon, sleep, or poll.\n- If work cannot continue, call `agentteams_block_subtask` with the same exact `subtask_id`, a\n  concrete reason, and available evidence. After a successful Block, stop and wait for new input.\n  The Leader owns further coordination and Human consultation; do not independently solicit other\n  people. After new input, re-read the Subtask and continue only when it is `in_progress`.\n\n## Task and Subtask files\n\nTask file operations are covered by this Skill; `file-sharing` is only for non-Task Team files.\n\n- Use `agentteams_list_task_files` and `agentteams_read_task_file` for parent Task inputs.\n- Use `agentteams_list_subtask_files` and `agentteams_read_subtask_file` for Subtask-owned files.\n  For a dependency's file, pass the owning dependency Subtask ID, not your assigned Subtask ID.\n- Determine the owner from the authoritative Task, Subtask, Result, or file-list response. Keep a\n  `fileRef` in the same owning space; do not construct it from an ID, path, URL, or example.\n- When ownership is absent or ambiguous, block the assigned Subtask and report the mismatch;\n  do not try both Task and Subtask spaces.\n\nWhen a file-list response contains `nextCursor`, request the next page in the same owning space\nuntil the cursor is absent or the required file has been found. Do not treat the first page as a\ncomplete listing.\n\nLocal staging paths are not durable and may disappear after restart. Persist required outputs with\n`agentteams_write_subtask_file` or `agentteams_write_subtask_file_from_path`, then retain the exact\nfile reference returned by Task Service. A local path is not a `fileRef`, and uploading a file does\nnot submit a Result. Treat `ok: true` as confirmation of that operation. Do not list, re-read, hash,\nor compare state solely to reconfirm a successful upload or write. Explicitly required content\nchecks still apply.\n\nThe Worker Prompt provides the collaboration workspace root. Keep all work for the current\nassignment under its exact Subtask ID: use `shared\u002Fsubtasks\u002F{subTaskId}\u002Finput\u002F` for parent Task or\ndependency inputs and `shared\u002Fsubtasks\u002F{subTaskId}\u002Fresult\u002F` for outputs. Do not stage Worker files\nunder `shared\u002Ftasks\u002F{taskId}\u002F`; that layout is owned by a Task Leader and would be shared by\nconcurrent Subtasks of the same Task. These are local workspace directories, not Team filesync\npaths or Task Service `fileRef` values. Collaboration file tools accept paths relative to the\nworkspace root.\n\nFor binary or large files, or content that should not pass through the model, use\n`agentteams_download_task_file` or `agentteams_download_subtask_file`. The destination must be inside\nthe configured collaboration workspace. The SDK obtains and consumes the short-lived download URL\ninternally; it is not returned to you. Use overwrite only after confirming that replacing the local\ndestination is intended.\n\nBefore every upload, inspect the selected content for secrets, credentials, tokens, authorization\nheaders, signed URLs, private keys, or passwords. Remove prohibited material through an authorized\nredaction or stop the upload. Pass the exact verified assignment ID as `subtask_id` to every Subtask\nfile write and Result submission. Include only required durable Subtask references in the Result.\n\nOn file-operation failure, use the structured error to choose the next step. Check ownership or\npaths only when the error points to them; do not guess a different reference or change credentials.\nDo not repeat an unchanged non-retryable operation. Report success only after `ok: true`.\n\n## Submit and stop\n\nUse the verification pass above to check the assigned outcome; do not create extra evidence just\nto restate acceptance criteria. Call `agentteams_submit_subtask_result` with the exact `subtask_id`\nand references returned by successful uploads. Use at most three short sentences for the summary:\nwhat was delivered, which required checks passed, and any material limitation. Do not repeat file\ncontents or enumerate acceptance criteria unless one is unmet or requires qualification.\nA successful submission moves the work to review; reply with one short sentence that it awaits\nreview, then stop. Do not poll or submit another Result while it is `pending_review`.\n\nIf review requests a revision, re-read the same Subtask and Result history, address only the stated\ngaps, and submit a new Result for the same Subtask. Do not call a Subtask resume operation; the\nparent Task Leader owns resume. Do not create or manage a top-level Task, reassign a Subtask, or\napprove your own Result. Stop retrying errors marked as non-retryable.\n","# Task execution\n\nYou are an AgentCore Worker. Task Service state is authoritative.\n\n## Reload and verify\n\n1. Copy the exact Subtask ID from the current notification or explicit user request, then pass it as\n   `subtask_id` to `agentteams_get_subtask`. Never infer an ID from a title or generate one. Treat the\n   returned object as the current assignment.\n2. Copy its exact `taskId`, then call `agentteams_get_task` for the parent Task. Do not derive the\n   parent ID from a message or local path.\n3. Verify the Subtask status, `assignedTo`, description, metadata, dependencies, and any referenced\n   specification. Inspect prior Results when resuming work or handling a revision. Use\n   `agentteams_list_task_events` only for history; current Task and Subtask objects remain\n   authoritative.\n4. When the assignee identity must be checked, call `agentteams_get_team_context` and compare\n   `assignedTo.userId` with this Worker's `member.matrixUserId`. Do not infer identity from names or\n   conversation text.\n5. Use the parent Task's `metadata.language` for acknowledgements, progress, blockers, Result\n   summaries, and revisions. If it is absent, consistently use the actionable request's language.\n\nFor every exact Subtask ID in `dependsOn`, call `agentteams_list_results` with the parent `task_id`\nand that dependency `subtask_id`. Do not execute until every dependency has a Result whose `status`\nis `accepted`. Use only the exact file references from that accepted Result when dependency output\nis required; do not infer completion from Subtask status, messages, events, or local files.\nDo not independently revalidate accepted dependency Results or parent Task facts; consume only\nwhat the current assignment needs.\n\nWhen a Subtask references a specification or input file, determine whether the reference belongs\nto the parent Task or the assigned Subtask before reading it. Use only the matching file tools; do\nnot try both spaces or rewrite the reference when ownership is ambiguous.\n\n## Follow the authoritative state\n\n- `assigned`: call `agentteams_ack_subtask` with the exact `subtask_id` exactly once before\n  substantive work or external side effects.\n- `in_progress`: continue without acknowledging again.\n- `blocked`, `pending_review`, `completed`, or `cancelled`: stop and wait for new input or an\n  authoritative state change.\n\nIf acknowledgement fails or its outcome is uncertain, re-read the Subtask once. Continue only when\nit is still assigned to you and is `in_progress`. If you have confirmed the Subtask is assigned to\nyou but information needed to execute it is missing, acknowledge it first and then block it.\nIf assignment is stale or does not match your identity, stop without writing.\n\n## Execute and report\n\n- Satisfy the Subtask description, expected outcome, dependencies, and every acceptance criterion\n  present in its metadata or parent Task. Perform every explicitly required test or live check.\n- Make one deliberate verification pass with the explicitly required checks and those necessary\n  to establish the requested artifact's usability. Once these pass, do not add exploratory tests,\n  repeated checks, hashes, or separate self-check artifacts. Explanatory or summary work does not\n  imply an execution check unless requested.\n- Keep deliverables no more detailed than requested. Do not add internal IDs, acceptance tables,\n  repeated calculations, or process narration unless the deliverable requires them.\n- Report progress only when it adds useful evidence; a heartbeat is not progress evidence.\n- If work continues for roughly 10 minutes since the latest successful acknowledgement, useful\n  progress update, or heartbeat, re-read the Subtask. Call `agentteams_heartbeat_subtask` with the\n  same exact `subtask_id` only when it remains assigned to you and `in_progress`. Do not start a\n  heartbeat daemon, sleep, or poll.\n- If work cannot continue, call `agentteams_block_subtask` with the same exact `subtask_id`, a\n  concrete reason, and available evidence. After a successful Block, stop and wait for new input.\n  The Leader owns further coordination and Human consultation; do not independently solicit other\n  people. After new input, re-read the Subtask and continue only when it is `in_progress`.\n\n## Task and Subtask files\n\nTask file operations are covered by this Skill; `file-sharing` is only for non-Task Team files.\n\n- Use `agentteams_list_task_files` and `agentteams_read_task_file` for parent Task inputs.\n- Use `agentteams_list_subtask_files` and `agentteams_read_subtask_file` for Subtask-owned files.\n  For a dependency's file, pass the owning dependency Subtask ID, not your assigned Subtask ID.\n- Determine the owner from the authoritative Task, Subtask, Result, or file-list response. Keep a\n  `fileRef` in the same owning space; do not construct it from an ID, path, URL, or example.\n- When ownership is absent or ambiguous, block the assigned Subtask and report the mismatch;\n  do not try both Task and Subtask spaces.\n\nWhen a file-list response contains `nextCursor`, request the next page in the same owning space\nuntil the cursor is absent or the required file has been found. Do not treat the first page as a\ncomplete listing.\n\nLocal staging paths are not durable and may disappear after restart. Persist required outputs with\n`agentteams_write_subtask_file` or `agentteams_write_subtask_file_from_path`, then retain the exact\nfile reference returned by Task Service. A local path is not a `fileRef`, and uploading a file does\nnot submit a Result. Treat `ok: true` as confirmation of that operation. Do not list, re-read, hash,\nor compare state solely to reconfirm a successful upload or write. Explicitly required content\nchecks still apply.\n\nThe Worker Prompt provides the collaboration workspace root. Keep all work for the current\nassignment under its exact Subtask ID: use `shared\u002Fsubtasks\u002F{subTaskId}\u002Finput\u002F` for parent Task or\ndependency inputs and `shared\u002Fsubtasks\u002F{subTaskId}\u002Fresult\u002F` for outputs. Do not stage Worker files\nunder `shared\u002Ftasks\u002F{taskId}\u002F`; that layout is owned by a Task Leader and would be shared by\nconcurrent Subtasks of the same Task. These are local workspace directories, not Team filesync\npaths or Task Service `fileRef` values. Collaboration file tools accept paths relative to the\nworkspace root.\n\nFor binary or large files, or content that should not pass through the model, use\n`agentteams_download_task_file` or `agentteams_download_subtask_file`. The destination must be inside\nthe configured collaboration workspace. The SDK obtains and consumes the short-lived download URL\ninternally; it is not returned to you. Use overwrite only after confirming that replacing the local\ndestination is intended.\n\nBefore every upload, inspect the selected content for secrets, credentials, tokens, authorization\nheaders, signed URLs, private keys, or passwords. Remove prohibited material through an authorized\nredaction or stop the upload. Pass the exact verified assignment ID as `subtask_id` to every Subtask\nfile write and Result submission. Include only required durable Subtask references in the Result.\n\nOn file-operation failure, use the structured error to choose the next step. Check ownership or\npaths only when the error points to them; do not guess a different reference or change credentials.\nDo not repeat an unchanged non-retryable operation. Report success only after `ok: true`.\n\n## Submit and stop\n\nUse the verification pass above to check the assigned outcome; do not create extra evidence just\nto restate acceptance criteria. Call `agentteams_submit_subtask_result` with the exact `subtask_id`\nand references returned by successful uploads. Use at most three short sentences for the summary:\nwhat was delivered, which required checks passed, and any material limitation. Do not repeat file\ncontents or enumerate acceptance criteria unless one is unmet or requires qualification.\nA successful submission moves the work to review; reply with one short sentence that it awaits\nreview, then stop. Do not poll or submit another Result while it is `pending_review`.\n\nIf review requests a revision, re-read the same Subtask and Result history, address only the stated\ngaps, and submit a new Result for the same Subtask. Do not call a Subtask resume operation; the\nparent Task Leader owns resume. Do not create or manage a top-level Task, reassign a Subtask, or\napprove your own Result. Stop retrying errors marked as non-retryable.\n",{"name":5,"version":24,"description":7},"1","main","a7a496e18ee243332d0518cdf6ab10ac8665dcc8","packages\u002Fcollaboration\u002Fworker-skills\u002Ftask-execution","https:\u002F\u002Fgithub.com\u002Fai-agentcore\u002Fagentcore-sdk-nodejs","https:\u002F\u002Fwww.aliyun.com\u002Fproduct\u002Fagentcore","https:\u002F\u002Fgithub.com\u002Fai-agentcore","2026-11-03 09:59:55",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add ai-agentcore\u002Fagentcore-sdk-nodejs --skill task-execution",{"tool":38,"command":39},"bunx","bunx skills add ai-agentcore\u002Fagentcore-sdk-nodejs --skill task-execution",{"tool":41,"command":42},"pnpm","pnpm dlx skills add ai-agentcore\u002Fagentcore-sdk-nodejs --skill task-execution","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fai-agentcore\u002Fagentcore-sdk-nodejs\u002Fblob\u002Fmain\u002Fpackages\u002Fcollaboration\u002Fworker-skills\u002Ftask-execution\u002FSKILL.md",true,[],[47,54,61],{"skillNo":48,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":51,"categories":52,"keywords":53,"publishedAt":19},"01M7CBPZE73YSBZ2MV5EVBYWD6","room-files","Download chat attachments and deliver local files or images to the current Matrix room.","https:\u002F\u002Fgithub.com\u002Fai-agentcore\u002Fagentcore-sdk-nodejs\u002Fblob\u002Fmain\u002Fpackages\u002Fcollaboration\u002Fworker-skills\u002Froom-files\u002FSKILL.md",[],[],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":58,"categories":59,"keywords":60,"publishedAt":19},"01M7CBPXVAW78CQ32TJTP9AY7E","file-sharing","Synchronize non-Task Team files; Task and Subtask files belong to task-execution.","https:\u002F\u002Fgithub.com\u002Fai-agentcore\u002Fagentcore-sdk-nodejs\u002Fblob\u002Fmain\u002Fpackages\u002Fcollaboration\u002Fworker-skills\u002Ffile-sharing\u002FSKILL.md",[],[],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":19},"01M7CBPWWFEZRWKFAAGRTC6HTH","greeting","用简短中文介绍助手可以如何帮助用户。","https:\u002F\u002Fgithub.com\u002Fai-agentcore\u002Fagentcore-sdk-nodejs\u002Fblob\u002Fmain\u002Fexamples\u002Fskills\u002Fgreeting\u002FSKILL.md",[],[]]