dme-skills
Plan work in Linear. Build one work issue. Hand it off with proof.
dme-skills is a public Agent Skills pack for a
consistent plan → build → review workflow. It keeps Linear, branches, PRs, estimates, code links, and
full session transcripts in sync.
| Command | Result |
|---|---|
/linearthis <idea> |
Shapes an idea into one standalone work issue, or an epic with work issues. |
/storythis |
Turns finished design/spec work into code-verified Linear work for developers. |
/ccthis <issue-or-epic> |
Builds one work issue on one branch + one PR. Given an epic, it picks one child and tells you which. |
/chonchi |
Hands the current work issue to review with its summary, actuals, code link, and full transcript. |
Commands are bare: use /ccthis, not /dme:ccthis.
Install
git clone https://github.com/jjdmev2/dme-skills ~/.claude/skills/dme-skills
~/.claude/skills/dme-skills/setup
Then:
- Run
/mcpin Claude Code and authenticate to Linear. - Make sure
nodeis installed; it powers transcript export. - Install gstack for the full workflow. Missing gstack skills fall back to native
git/ghwhere possible.
The repo is public, so cloning needs no GitHub authentication. The skills work in Claude Code CLI, the VS Code Claude Code extension, and Conductor.
Run the workflow
1. Plan
/linearthis add branded share links
/linearthis checks existing Linear work, reads the relevant code, and asks only the decisions that
change scope. It creates either:
- one standalone work issue when the outcome is one package; or
- one epic with multiple work issues when the outcome needs several packages.
Finished a design first? Run /storythis to create the design-to-development handoff instead.
2. Build one work issue
/ccthis KIS-160
/ccthis builds one work issue per run, keeps Linear updated, and ships one branch + one PR. It never
completes untouched siblings or the parent epic.
Building is a conversation: review the work, test it, and iterate before handoff.
3. Hand it off
/chonchi
When the current work issue is finished and tested, /chonchi:
- Moves the work issue to In Review.
- Attaches the full session transcript and a direct PR, branch, or commit link.
- Creates or updates one session summary with estimate, session actual, and cumulative actual.
- Applies
chonchionly after the transcript and code link succeed. - Refreshes one roll-up comment on the parent epic, if any.
This split is deliberate: exporting during /ccthis would miss later feedback and revisions.
Retry and reopen safely
- Same-session retry: reuses the marked comment, deterministic transcript attachment, and identical code link. The session's hours count once.
- Later session: adds one new session record and adds that session's actual once to the cumulative total.
- Reopened work:
/ccthisremoves only the stalechonchilabel, preserves other labels, and returns an unstarted or review-stage parent epic to In Progress.
Rules that keep Linear correct
- An epic is one coordinated product outcome. A work issue is one ownable, reviewable, verifiable package.
- A code work issue never spans repositories. One repository may contain several work issues.
- The work issue owns its branch, PR, estimate, actuals, state, and handoff. The epic only aggregates its children; one child's data never overwrites the epic.
- A
chonchilabel means the transcript and code link both landed. A child counts as handed off only when it also has a review/completed state. - The epic moves to In Review only when every child is handed off. A partial handoff never promotes the epic to In Review and never changes untouched siblings.
Reference
Update
/dme-upgrade
This pulls the latest version and re-runs setup. Manual update:
cd ~/.claude/skills/dme-skills && git pull && ./setup
This repository previously lived at dmenetwork/dme-cc. Existing clones keep working through GitHub's
redirect and the skills' legacy-path fallback. New installs should use jjdmev2/dme-skills.
Transcript export and security
Claude Code stores each session as JSONL. The installed session hook records the active transcript path;
/chonchi converts it to Markdown, uploads it to Linear, and attaches the current GitHub link. You do not
need to run /export manually.
Transcripts can contain secrets printed in the terminal, including tokens, .env values, keys, and signed
URLs. Review and redact the transcript before upload unless the Linear workspace and issue are private.
Layout
linearthis/SKILL.md # plan: idea → standalone work issue or epic + work issues
storythis/SKILL.md # design handoff → project, design issue, and build stories
ccthis/SKILL.md # build: one work issue → one branch + one PR
chonchi/SKILL.md # review: transcript + code link + actuals + epic roll-up
dme-upgrade/SKILL.md # update the pack and re-run setup
scripts/
session-hook.sh # records the active transcript path
transcript-to-md.mjs # converts Claude JSONL to Markdown
gh-links.sh # finds the current repo, branch, commit, and PR
setup # installs the hook and configures the Linear MCP
License
MIT — see LICENSE.
No comments yet
Be the first to share your take.