How do I install the Dstack Implement skill?

Run npx skillmds@latest add robertderose/dstack-implement 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 Dstack Implement safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Dstack Implement?

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 Dstack Implement free to use?

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

Who published Dstack Implement?

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

Dstack Implement by robertderose — Agent Skill

Name: Dstack Implement
Author: robertderose

File contents

Implement

Use this skill only when explicitly invoked; that invocation activates the dStack workflow. Within an active workflow, Beads is the only next-work surface. Do not read a task list file or maintain a separate progress record.

Establish the worktree

Run:

dstack worktree --bead <feature-root>

Change into the returned worktree and verify the active branch matches the returned feat/<slug> branch. Do not run bd prime; this skill already supplies the active dStack workflow context.

Claim one task

bd ready \
  --parent <implementation-epic> \
  --label dstack:work:implementation \
  --claim \
  --json

Resolve the implementation epic from the formula step label before claiming. Read the claimed task, its direct blockers, referenced decisions, and only the source and documentation needed for that outcome. If no implementation task is ready, report native blockers; do not invent a dStack phase or claim the audit.

Implement the complete accepted outcome

Implement the smallest complete change satisfying the task. Keep code, tests, configuration, and current user/developer/future-agent documentation aligned in the same unit of work. Do not postpone documentation to a closeout phase.

When new product or architecture ambiguity appears, record it on the task and ask the user rather than guessing. Clear incidental defects may be fixed in scope. Significant separate work becomes a native Beads task linked with discovered-from.

Review the complete diff, stage only task-owned repository changes, and commit through dStack so the subject and footer come from the Bead:

dstack commit --bead <task-id> [--body <temporary-body>]

Do not hand-write a Conventional Commit subject. If subject generation fails, correct the task title or its dstack:commit:* / dstack:scope:* labels.

Validation is mandatory and uses the repository's hk contract:

dstack check task --bead <task-id>

Fix every reported error. When the task intentionally requires no repository change, add a durable Beads note containing exactly:

No repository change: <specific reason>

Then rerun the check. Close the task only after it passes:

bd close <task-id> --reason 'Accepted outcome implemented and validated'

Implement one task by default. Continue only when the user explicitly requests all ready work. Never calculate fan-in or advance the final audit yourself.

---name: dstack-implementdescription: Claim and implement native ready work with deterministic worktree, commit, evidence, and validation checks.---# ImplementUse this skill only when explicitly invoked; that invocation activates the dStack workflow. Within an active workflow,Beads is the only next-work surface. Do not read a task list file or maintain a separate progress record.## Establish the worktreeRun:```bashdstack worktree --bead <feature-root>```Change into the returned worktree and verify the active branch matches the returned `feat/<slug>` branch. Do not run`bd prime`; this skill already supplies the active dStack workflow context.## Claim one task```bashbd ready \ --parent <implementation-epic> \ --label dstack:work:implementation \ --claim \ --json```Resolve the implementation epic from the formula step label before claiming. Read the claimed task, its direct blockers,referenced decisions, and only the source and documentation needed for that outcome. If no implementation task is ready,report native blockers; do not invent a dStack phase or claim the audit.## Implement the complete accepted outcomeImplement the smallest complete change satisfying the task. Keep code, tests, configuration, and currentuser/developer/future-agent documentation aligned in the same unit of work. Do not postpone documentation to a closeoutphase.When new product or architecture ambiguity appears, record it on the task and ask the user rather than guessing. Clearincidental defects may be fixed in scope. Significant separate work becomes a native Beads task linked with`discovered-from`.Review the complete diff, stage only task-owned repository changes, and commit through dStack so the subject and footercome from the Bead:```bashdstack commit --bead <task-id> [--body <temporary-body>]```Do not hand-write a Conventional Commit subject. If subject generation fails, correct the task title or its`dstack:commit:*` / `dstack:scope:*` labels.Validation is mandatory and uses the repository's hk contract:```bashdstack check task --bead <task-id>```Fix every reported error. When the task intentionally requires no repository change, add a durable Beads note containingexactly:```textNo repository change: <specific reason>```Then rerun the check. Close the task only after it passes:```bashbd close <task-id> --reason 'Accepted outcome implemented and validated'```Implement one task by default. Continue only when the user explicitly requests all ready work. Never calculate fan-in oradvance the final audit yourself.

Frequently asked questions

Dstack Implement