agentsclimarketplace
Sign in

M plan roll

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

Autonomous "full send" of the m_plan pipeline — plan at large scope and implement, no questions. Forces the full 9-artifact plan, decides every blocker itself (recording assumptions), then chains straight into m_plan_implement and drives to done + deploy + verify without pausing. Use when you explicitly want the whole architect → build → deploy → verify cycle run hands-off on a substantial task. Heavier and far less interactive than /m_plan — reach for it only when you truly want zero questions.From its SKILL.md

Install
npx -y skills add mapuamap/denys-fast-mskills --skill m_plan_roll

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

3 things to look at

  • skips confirmationTells the agent to proceed without asking first, 4 times: "No `AskUserQuestion`, no blocker batch." and 3 more.
  • 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.
  • runs commandsInstructs the agent to run 3 commands, including `Run `m_plan` Phases 0–1` and 2 more.

SKILL.md

4.5 KB, ~1.0k tokens by cl100k_base, as published. Nobody here has run it

m_plan_roll

One-shot, hands-off pipeline: plan large → implement → deploy → verify, no questions. A thin wrapper over m_plan (Phases 0–1) and m_plan_implement (execution) that swaps the interactive parts for autonomous defaults and maximum coverage.

This trades interactivity for autonomy. It still cannot fake verification and still obeys the hard safety gates below — those are the only things that can halt or block the run.

Inputs

  • $ARGUMENTS: the task. If empty, ask once for the task, then stop asking.

How it differs from /m_plan

m_plan asks blockers and right-sizes. m_plan_roll does neither — it assumes the most thorough path and decides everything itself.

Phase 1 — Plan (autonomous, large)

Run m_plan Phases 0–1 with these overrides:

  • No AskUserQuestion, no blocker batch. For every blocker in m_plan's BLOCKERS.md, pick the safest thorough default yourself. Record each non-obvious decision under a ## Assumptions heading in 01_architecture.md so the run stays auditable.
  • Force size = large → write all 9 artifacts (01–09), maximum coverage (08 included).
  • Gitignore: append .m_plan/ to .gitignore by default (no prompt); note it in the run log.
  • No Approve pause. Do not stop for Approve? (yes / edit / stop) — go straight to Phase 2.
  • Still obey m_plan's artifact rules: a per-step check in every 05 step, build/test commands live only in 07, fill every <…> placeholder, no secrets / real tokens / prod hostnames.

Phase 2 — Implement (autonomous)

Immediately invoke m_plan_implement and run its Phase B to completion, with these overrides:

  • Auto-answer its prompts: Resume the plan just written; Ready? = yes; no slug-selection question (use the new plan).
  • Deploy runs automatically once every non-deploy V-* is [x]/[~] (treat it as if the user typed deploy) — against the real declared target only, per 06 or the discovered deploy path.
  • If a non-deploy V-* is itself [!] blocked (e.g., the target or a required check is unreachable), the auto-deploy precondition isn't met — mark the dependent V-DEPLOY-* / V-SMOKE-* [!] too, finish everything else, and report PARTIAL. Don't force a deploy and don't hang waiting.
  • Walk steps, flip V-*, log deviations immediately, run the final V-* sweep, then emit the three-section report.

Hard safety gates (still apply — never bypassed)

Autonomy means don't ask for preferences. It does not mean skip safety:

  • Never fake verification. 09_verification.md still decides done. Local build/unit/dev-server checks never satisfy V-DEPLOY-* / V-SMOKE-* / V-E2E-* (the deploy reality invariant).
  • Can't access a required check (no real deploy path, or the browser MCP / credentials / URL for a check are missing): do not stop the whole run and do not fake it — mark that V-* [!] blocked, log a deviation, and keep going with everything else. Surface it in the final report.
  • Destructive actions on production / shared services (restarts, migrations, replacing containers, data deletion, reboots): apply the repo/user destructive-action policy. If that policy requires confirmation, this is the one place the run pauses — stop at that specific action and report it rather than doing something irreversible unprompted.
  • No secrets / real tokens / prod hostnames in any artifact or commit.

Phase 3 — Report

Emit m_plan_implement's three-section report (✅ Done / 🔀 Worked around or changed / ❌ Not done), then append:

  • Assumptions made: every autonomous decision from Phase 1 (copy from 01's ## Assumptions).
  • Halted on: any destructive-action gate that paused the run (or "none").
  • Final: DONE | PARTIAL | BLOCKED. If any V-* is [ ]/[!], it is not DONE — never close by lying.

Files in this skill

  • SKILL.md — this file (orchestrator; execution semantics live in m_plan / m_plan_implement)
  • evals/evals.json — test cases

What ships with it: 1 file

1.7 KB alongside SKILL.md

evals/

Gives 0 of the 12 instructions most plan spec skills give in ~1.0k tokens

Counted across 1,360 of the 2,617 authors here whose files we hold, read 2026-09-06

  • Ask one question at a timein 73 of 1360
  • Write the spec using the templatein 22 of 1360
  • Ask clarifying questions if neededin 19 of 1360, across 18 files
  • Wait for user confirmation before proceedingin 19 of 1360
  • Save plans to the plans directoryin 17 of 1360, across 13 files
  • Check for product marketing context firstin 16 of 1360, across 5 files
  • Read the plan file completelyin 16 of 1360
  • Order tasks by dependencyin 16 of 1360
  • Gather context from the conversationin 15 of 1360, across 9 files
  • Explore the codebase instead of askingin 15 of 1360, across 13 files
  • Wait for explicit user approvalin 14 of 1360, across 13 files
  • Quiz the user on the breakdownin 13 of 1360, across 7 files

Said here and by no other author read

  • Log any deviations immediately in the verification document
  • Emit a three-section report at the end of execution
  • Run m_plan phases zero and one
  • Pick the safest thorough default blocker
  • Record each non-obvious decision under assumptions
  • Write all nine planning artifacts

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. 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.