agentleFS
rumoiirumoii/rumo-engineering-skills/skills/rumo-plan/SKILL.md

rumo-plan

Skill7 starsChanged 14 days agoMITView on GitHub

Use when turning an agreed multi-step or cross-module task into a short spec and verifiable task breakdown before implementation, when executing an agreed plan task by task, or when creating or maintaining a durable engineering decision record (ADR) for a cross-module contract, data or wire format, permission model or security policy, lifecycle rule, deployment, migration, or rollback. Do not use for single-file fixes or mechanical edits, and do not write a duplicate record for a choice an existing record already owns (update or supersede it instead); settle unclear goals with rumo-clarify first.

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 8 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before a task, re-read the files it touches and the check it names.

    line 35
  2. Distinctive:

    After the task, run that check.

    line 36
  3. Distinctive:

    When a check fails, stay on that task until it passes or the plan needs revising.

    line 37
  4. Distinctive:

    Read references/decision-records.md before writing or moving a record.

    line 45
  5. Distinctive:

    Search for an existing record that owns the same mechanism.

    line 47
  6. Distinctive:

    Follow the repository's existing location, naming, numbering, and language.

    line 48
  7. Distinctive:

    Do not mark a record implemented because code was drafted or unit tests passed.

    line 49
  8. Distinctive:

    Do not create decision directories or policies during a review-only request.

    line 51

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Rumo Plan

Make the work and its reasons explicit before they become code.

Modes

ModeUse whenOutput
PlanThe task spans several steps, modules, or sessionsA short spec plus ordered, verifiable tasks
ExecuteAn agreed plan is ready to implementTasks done in order, each with its check result
Decision recordA material choice must outlive this conversation, or an existing record needs updatingA decision record in the repository's convention

Unclear goals or acceptance criteria go to rumo-clarify first. A decision record can be maintained even for a small change when it owns the affected mechanism.

Plan Mode

  1. Inspect the real entry points, owners, tests, and repository instructions for the affected area.
  2. Write the spec in a few lines each: goal, non-goals, constraints, acceptance criteria, and open decisions with recommended answers. When more than one design is viable and the choice changes an interface, stored data, or the amount of work, list two or three options with their trade-offs and a recommendation, and break the work down only after the user picks one. Options must differ in where the change lives, how data flows, or what it costs; the same design with another parameter or library is one option. Before recommending, check each option for these design smells: a layer that only forwards calls and hides nothing; one piece of knowledge, such as a file format or business rule, spread across several modules; modules split by execution order (read, process, write) instead of by the knowledge each owns; one piece of state written by more than one module; two supported ways to do the same task; and a list that must be kept in step by hand in several places. When the user wants another model's design to compare, hand it a read-only design brief through rumo-agent-collaboration and compare its draft with yours.
  3. Break the work into tasks. Each task names the files or modules it touches, the check that proves it done, and what it depends on. Prefer thin end-to-end slices over layer-by-layer batches.
  4. Mark tasks that need separate authorization: commits, tags, pushes, deployments, database writes, deletions, credential changes, dependency installs, and paid services.
  5. Show the plan. When the user already authorized this scope and the plan adds no new decision, continue; otherwise wait for approval of that exact version, pausing only the tasks that depend on the open decision or authorization. A clarifying answer revises the plan; it is not approval.
  6. While implementing, stop and revise the plan when a task turns out wrong or a hidden requirement appears; do not silently drift.

Save the plan where the user or repository wants it. Without a convention, keep it in the conversation rather than inventing a directory.

Execute Mode

Work an agreed plan one task at a time, in dependency order.

  1. Before a task, re-read the files it touches and the check it names.
  2. After the task, run that check. Mark the task done only when the check passes, and record the command and result.
  3. When a check fails, stay on that task until it passes or the plan needs revising; never start the next task on top of a failing one.
  4. Report per task or small batch, and end with one line per task: done with its evidence, blocked with the reason, or not started.
  5. When the plan lives in a file and the work spans several steps or sessions, keep a "Decisions and dead ends" section at its end: append one line for each fork, failure, or change of direction, with the date, the decision, why, where the evidence is, and the result. Record approaches that were tried and abandoned with the reason, so the next session does not try them again. Skip routine steps, and never edit or delete an earlier line; a reversed decision gets a new line.

Run tasks in parallel or in separate worktrees only after the user agrees, and never run two tasks in parallel when they touch the same files. Offer it once per plan, in one line naming the tasks, who would take each, and why, when at least two tasks can run together under references/parallel-execution.md, none of them is a few-line edit, and a delegation route exists: a configured delegate lane, or a session running inside Orca. Read that reference before offering or running them.

Decision Record Mode

Read references/decision-records.md before writing or moving a record.

  • Search for an existing record that owns the same mechanism. Update it when only facts, paths, or evidence changed; write a new record that links the old one when the decision itself changed.
  • Follow the repository's existing location, naming, numbering, and language. Without one, propose the default layout in the reference and ask before creating it.
  • Do not mark a record implemented because code was drafted or unit tests passed; state the source, artifact, deployment, or field evidence that exists.
  • A decision record never authorizes database writes, deployments, or other controlled actions.
  • Do not create decision directories or policies during a review-only request.

Report

For plans: the spec, the task list with checks, items needing separate authorization, and the approval status. For execution: each task's status with its check evidence. For records: status, path, superseded records, and evidence boundaries.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.