agentsclimarketplace

Trtc push

Skill Tencent-RTC/agent-skills/skills/trtc-push

Use for TIMPush (Tencent Cloud IM Push / 腾讯云即时通信推送) developer assistance — Android and iOS integration and troubleshooting via trtc-push-mcp, plus Flutter, uni-app, HarmonyOS, server API, badge, console, and product-limit routing. Use when the user is integrating or debugging TIMPush itself (e.g. 接入 TIMPush, 帮我集成 timpush, 集成 TIMPush, 集成腾讯云 push, 接入腾讯云 push, 集成腾讯云离线推送, 接入腾讯云离线推送, 即时通信推送, 离线推送接入, TIMPush 离线推送, 腾讯云离线推送, 腾讯云 push, integrate TIMPush, integrate Tencent Cloud push, integrate Tencent Cloud offline push, setup Tencent Cloud push, registerPush failed, 800006, FCM unavailable with TIMPush, APNs, businessID). Entered via trtc dispatcher pre-gate when shipped inside trtc-agent-skills. Do not use for maintaining/debugging this skill or MCP reporting tooling, Conference/TUIRoom/直播间/口语陪练/AI客服 主导且无 TIMPush 信号的场景, or unrelated generic push work with no TIMPush signal.From its SKILL.md

Install
npx -y skills add Tencent-RTC/agent-skills --skill trtc-push

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 15 stars15 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

7.1 KB, as published. Nobody here has run it

TIMPush 开发者助手

渐进披露

  • 每 turn 先读 issues/ROUTER.json 做轻量路由;不要先读整个 issues/。
  • 进入 Android / iOS / Flutter / UniApp workflow 前,先读一次 references/hard-rules.md。
  • 写 Gradle / Application / Podfile / AppDelegate / registerPush 代码前,再读 references/timpush-sdk-api.md 和 references/code-templates.md。
  • 路由命中非 workflow 知识时,只读命中的 1 个 target。
  • 启动字段、abandon_workflow 等低频协议细节见 references/workflow-protocol.md。

核心原则

本 Skill 不含可由 LLM 自行解释的流程正文。阶段、schema、失败路由都由 trtc-push-mcp 的 workflow engine 决定。

你的工作:按 engine 下发的当前 prompt 执行单步 → complete_workflow_step 提交结构化 output → 下一步只能来自引擎返回值。禁止自己编流程、跳步、或用 markdown 手册替代 engine。

每 turn(按顺序)

  1. 读 issues/ROUTER.json 的 match_any / priority / platform,得到候选 target / workflow_id。路由是 best-effort;未命中不代表不能处理。
  2. 命中 platform=auto 的集成入口(如“集成 push / 接入离线推送”)→ 先调用 detect 系列工具判断项目平台,再选择 wizard-android / wizard-ios / wizard-flutter / wizard-uniapp。
  3. 命中 Android / iOS / Flutter / UniApp workflow_id → 读 references/hard-rules.md,进入 Workflow 循环。
  4. 需要解释或非 workflow 知识 → 只读路由命中的 1 个 target。
  5. 无命中 → 按 fallback.question 问 1 个澄清问题。
  6. 意图不清 → list_workflows,把 trigger_phrases 复述给用户澄清。

特殊分支:

  • 用户已明确是 Android / iOS TIMPush 排障但缺厂商/错误码 → 进对应 troubleshoot-* stage-0 收集症状与证据,不要自行下结论
  • list_workflows 未返回预期 workflow → 提示用户重启 IDE / 确认 trtc-push-mcp MCP 配置,不要退回 markdown 手册执行。

Workflow 循环

list_workflows(可选,看候选)
  → get_workflow_state(产品使用每 turn 带 user_prompt;调试不传)
  → 按返回的 prompt / required_tool_calls / completion_schema 执行
  → complete_workflow_step { run_id, step, output }(首选)
  → 按 status 继续或停止

run_id 是 workflow 主路径状态句柄;state_token 只作为兼容 / 跨进程快照。后续 get_workflow_state、complete_workflow_step、abandon_workflow 都优先传 上一轮返回的 run_id。state_token 必须视为 opaque:禁止手改、拼接、解码后重组、 局部复制;如果只剩 state_token,只能原样回传。

Prompt 上报(MCP 自动,勿再跑 shell)

  • 产品使用时,每轮向 get_workflow_state 传 user_prompt;工具调试或无法判断时不传,质量事件和本地脱敏由 MCP 统一处理。
  • 不得把 user_prompt 放入 context,也不得执行 --prompt-stdin / --log-stdin;避免原文进入 state_token 或额外 shell 链路。

complete_workflow_step 返回值

  • advanced → 用返回的 run_id 进入 next_state,执行新 prompt(state_token 仅兼容)。
  • failure_retry → 按 error_code + hints_for_llm 修,再次提交。
  • schema_violation → 按 errors 改 output,同一 run_id/state 再提交,不要前进。
  • advanced_to_failure → 告诉用户原因,stop。
  • invalid_token / unknown_run → 先用 run_id 恢复;没有可用 run_id 时重启同一 workflow。不得手工修 token。若重启仍失败或 MCP 不可用,进入“受控 fallback”。

绝不自己决定「下一步该做什么」或越过当前 state——next_state 只能来自引擎。

abandon_workflow 每个 run 最多一次,只在用户明确放弃/换话题,或同一 run_id 反复失败且无路可走时调用。正常收尾、单次可恢复失败、只想暂停时不要调用。

执行时:MCP 返回的 prompt 优先于训练记忆 / 常识。

受控 fallback(workflow 异常时)

目标是继续帮用户快速接入,不是机械报错。只有在 workflow 连续无法恢复(如 MCP 不可用、run_id/state_token 都无法继续、同一阶段反复工具异常)时进入 fallback。

fallback 必须遵守:

  • 先告知用户:主 workflow 中断,正在按同一套 TIMPush 约束走备份路径。
  • 继续执行 references/hard-rules.md;写代码前仍读 references/timpush-sdk-api.md 和 references/code-templates.md。
  • 尽量继续使用 MCP 原子工具(检测、版本、校验、写 local 配置),不要自由发挥。
  • 凭据安全(iOS:Credentials 源文件 + gitignore,禁止主 Info.plist;Android:local.properties)、registerPush 后打印 registrationID=、iOS 不自动改 .pbxproj Capability 等红线不因 fallback 放宽。
  • 收尾必须列出:已完成项、未完成 workflow stage / 人工 checklist、验证结果、剩余风险。

知识库用法

  • 写代码前按需读 references/timpush-sdk-api.md、references/code-templates.md。
  • issues/:先 ROUTER.json;cards/ = 固定问题卡,flows/ = 分支排查。不要通读整个 issues/。
  • Android / iOS / Flutter / UniApp 排障仍必须走 MCP workflow;issues/**/*.md 只能作知识与分支解释,不能替代 list_workflows → get_workflow_state → complete_workflow_step。
  • HarmonyOS / RN / 产品 FAQ:有对应 workflow 前走知识文档;不要把它们塞进其它平台 workflow。

已知限制(避免误用)

  • MCP workflow 当前覆盖以 list_workflows 返回为准;RN / HarmonyOS 仍主要通过 issues/ 做排查引导。
  • Flutter 工程请走 wizard-flutter;UniApp 请走 wizard-uniapp,不要误跑纯 wizard-android / wizard-ios(除非用户明确只要改某一原生子工程)。
  • workflow engine 仍是 alpha;required_tool_calls 目前由 AI 自报,尚未升到 proof token。

What ships with it: 20 files

77.4 KB alongside SKILL.md

Gives 0 of the 12 instructions most debug triage skills give

Counted across 1,020 of the 1,639 authors here whose files we hold, read 2026-09-06

  • Find root cause before attempting any fixin 134 of 1020, across 118 files
  • Create a failing test case before implementing a fixin 109 of 1020, across 95 files
  • Read error messages and stack traces completelyin 102 of 1020, across 88 files
  • Reproduce the issue consistently before investigatingin 90 of 1020, across 77 files
  • Make the smallest possible change to test a hypothesisin 90 of 1020, across 76 files
  • Trace data flow backward to find the sourcein 84 of 1020, across 70 files
  • Form a single hypothesis before testingin 78 of 1020, across 64 files
  • Implement only one fix at a timein 76 of 1020, across 63 files
  • Question the architecture if three fixes failin 73 of 1020, across 59 files
  • Add diagnostic instrumentation at component boundariesin 68 of 1020, across 56 files
  • Compare broken code against working examplesin 68 of 1020, across 57 files
  • Write a regression test before applying the fixin 62 of 1020, across 55 files

Said here and by no other author read

  • Read ROUTER.json before reading the issues directory
  • Read hard-rules.md before entering platform workflows
  • Read SDK API and code templates before writing code
  • Execute steps provided by the workflow engine
  • Submit structured output via complete_workflow_step
  • Use the run_id provided by the engine for all steps

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once.

Keep looking

Skills are one crate of 613,297. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.