M go

Skill mapuamap/denys-fast-mskills/skills/m_go

Claude Code plugin: m_* toolkit (m_plan plan->execute->verify, deploy/infra commands, Playwright full-test loop, m_code quality framework)

Install

npx -y skills add mapuamap/denys-fast-mskills --skill m_go

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

1 stars1 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.

What its author says it does

Copied from the file, not written here

Resume the last saved work session — read the standing handoff file `.m_last_task.md` (written by /m_save) at the project root, restore context by reading the key files it names, then continue from its recorded "next step". Use at the start of a fresh session to pick up exactly where the previous one stopped.

SKILL.md

2.3 KB, as published. Nobody here has run it

m_go

Read .m_last_task.md from the project root (written earlier by /m_save), rebuild working context, and continue the work — not just summarize it.

Inputs

$ARGUMENTS (optional): adjustments to the resumed task ("skip the tests part", "focus on X first"). Apply them on top of the saved plan.

Procedure

Read the snapshot. Read .m_last_task.md at the project root.
- Missing → tell the user there is no saved task (/m_save hasn't been run here) and stop.
- Present in a parent directory but not here → mention it and ask which project they meant; don't guess.
Sanity-check freshness. Compare the snapshot's branch/commit line with git branch --show-current and git log -1 --oneline. If the repo has moved on (different branch, new commits, dirty tree not explained by the snapshot), say so in one line and account for it — the snapshot describes intent, the repo describes reality; reality wins.
Restore context. Read every file under "Key files" (relevant parts, not blindly whole files). If the snapshot links an .m_plan/<slug>/ run, open 05_step_plan.md and 09_verification.md — for a plan run, prefer resuming via /m_plan_implement (it owns the step-walking and V-* bookkeeping) and say you're doing so.
Brief the user (short). 3–5 lines: the task, what's done, and the next step you're about to take. This is a status line, not a report — then keep going without waiting, unless "Blockers / open questions" contains something only the user can answer (ask exactly that, then proceed).
Continue the work from "⏭ Next step", honoring "Decisions & context" (don't re-litigate settled decisions) and using "How to verify" to check your results.
Keep the snapshot current. When the resumed task completes, or the user switches to something else mid-run, update .m_last_task.md accordingly (an /m_save-style overwrite) so the file never describes stale work as pending.