How do I install the Tiangong Auto Research Workbuddy skill?

Run npx skillmds@latest add tiangong-ai/tiangong-auto-research-workbuddy in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Tiangong Auto Research Workbuddy safe to use?

This skill has not completed SkillMD's automated safety review yet. Independent scanners report: SkillSpector: PASS, Skill Scanner: PASS. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Tiangong Auto Research Workbuddy?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Tiangong Auto Research Workbuddy free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Tiangong Auto Research Workbuddy?

tiangong-ai (@tiangong-ai) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Tiangong Auto Research Workbuddy by tiangong-ai

Name: Tiangong Auto Research Workbuddy
Author: tiangong-ai

Tiangong Auto Research Workbuddy

Adapt a managed Tiangong Auto Research workflow to a native WorkBuddy or CodeBuddy producer running in Default Permission mode, especially when independent review needs the explicit sandbox-bridge transport. Use only as a thin router to the project-installed tiangong-auto-research Skill; do not use for unrelated WorkBuddy tasks or standalone searches.

tiangong-ai Updated Sep 22, 2026

File contents

Tiangong Auto Research for WorkBuddy

For feedback or Issue requests, load the sibling tiangong-auto-research/SKILL.md and follow its reporting route before workspace inspection. Its installed references/issue-reporting.md supplies the common templates and reporting rules.

Use the research workflow below only inside a user-selected managed Auto Research workspace. Inspect the exact workspace context first. If setup is absent, use the canonical Skill's setup reference; do not copy control files or invent a workspace.

Load the sibling project-installed tiangong-auto-research/SKILL.md, then read its references/sandboxed-ide.md and the other reference selected by the current stage. That canonical Skill owns the research workflow, schemas, budgets, evidence rules, and recovery behavior; this adapter does not duplicate them.

Run canonical CLI operations through this adapter's scripts/workbuddy_research_cli.sh. It selects only an explicit Node.js 24 runtime and then delegates to the sibling canonical resolver. Do not call WorkBuddy's ambient Node 22/npx after an EBADENGINE warning and do not suppress the engine requirement.

Keep WorkBuddy in Default Permission mode. Never choose Full Access, an unsandboxed-command escape hatch, dangerouslyDisableSandbox, dangerously-skip-permissions, excludedCommands, or silent transport fallback.

The workspace must identify the actual native producer as workbuddy or codebuddy. Use that same value for the packet's --host-agent; never impersonate Codex or Claude. The CLI must not launch the producer as a child.

For sandbox-bridge, require the owner to start the exact-version reviewer sidecar from a separate native terminal and a private state directory outside the workspace. From the IDE, verify research reviewer status, then run the explicitly cost-confirmed reviewer doctor. Stop on any structured bridge, signature, nonce, model, version, result-binding, or sandbox-policy failure.

After readiness passes, perform native producer stages in this current task and return every result through the canonical CLI packet. Independent review and mechanical closure remain CLI-controlled.

---name: tiangong-auto-research-workbuddydescription: Adapt a managed Tiangong Auto Research workflow to a native WorkBuddy or CodeBuddy producer running in Default Permission mode, especially when independent review needs the explicit sandbox-bridge transport. Use only as a thin router to the project-installed tiangong-auto-research Skill; do not use for unrelated WorkBuddy tasks or standalone searches.---# Tiangong Auto Research for WorkBuddyFor feedback or Issue requests, load the sibling `tiangong-auto-research/SKILL.md`and follow its reporting route before workspace inspection. Its installed`references/issue-reporting.md` supplies the common templates and reporting rules.Use the research workflow below only inside a user-selected managed Auto Research workspace.Inspect the exact workspace context first. If setup is absent, use the canonicalSkill's setup reference; do not copy control files or invent a workspace.Load the sibling project-installed `tiangong-auto-research/SKILL.md`, then readits `references/sandboxed-ide.md` and the other reference selected by thecurrent stage. That canonical Skill owns the research workflow, schemas,budgets, evidence rules, and recovery behavior; this adapter does not duplicatethem.Run canonical CLI operations through this adapter's`scripts/workbuddy_research_cli.sh`. It selects only an explicit Node.js 24runtime and then delegates to the sibling canonical resolver. Do not callWorkBuddy's ambient Node 22/npx after an `EBADENGINE` warning and do not suppressthe engine requirement.Keep WorkBuddy in Default Permission mode. Never choose Full Access, anunsandboxed-command escape hatch, `dangerouslyDisableSandbox`,`dangerously-skip-permissions`, `excludedCommands`, or silent transportfallback.The workspace must identify the actual native producer as `workbuddy` or`codebuddy`. Use that same value for the packet's `--host-agent`; neverimpersonate Codex or Claude. The CLI must not launch the producer as a child.For `sandbox-bridge`, require the owner to start the exact-version reviewersidecar from a separate native terminal and a private state directory outsidethe workspace. From the IDE, verify `research reviewer status`, then run theexplicitly cost-confirmed reviewer doctor. Stop on any structured bridge,signature, nonce, model, version, result-binding, or sandbox-policy failure.After readiness passes, perform native producer stages in this current task andreturn every result through the canonical CLI packet. Independent review andmechanical closure remain CLI-controlled.

Frequently asked questions