agentleFS
agentstationagentstation/skills/plans/SKILL.md

plans

Skill1 starsChanged 2 months agoMITView on GitHub

Use this skill to create, promote, execute, resume, or close a durable implementation plan that acts as a control plane with a status ledger, verifiable task criteria, and a goal block for autonomous execution. Use it when work spans several sessions or pull requests, changes architecture across seams, or must survive context compaction. Do not use it for a small change that one session delivers without coordination.

6 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 4 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read references/structure.md for the vocabulary, statuses, sections, ledger rules, and evidence rules.

    line 20
  2. Very specific:

    After each edit to an HTML plan or index, run scripts/plans check on it.

    line 29
  3. Very specific:

    When execution reveals an architecture problem, apply references/architecture.md.

    line 34
  4. Very specific:

    Before delivery or a release, read references/delivery.md for acceptance, compatibility, and rollout evidence.

    line 36
  5. Distinctive:

    Write plan prose with the technical-writing skill in developer mode.

    line 38
  6. Distinctive:

    Keep one status token per ledger cell, with the proof in the evidence cell.

    line 47
  7. Distinctive:

    When the commit policy permits, record each ledger transition in a plan commit right after its work commit.

    line 56
  8. Distinctive:

    Keep the plan a thin control plane.

    line 60

“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “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

Plans

A plan is one durable file that owns one outcome. The plan file, its status ledger, and the git worktree hold progress, so any agent resumes from them without chat history. Compaction erases conversation state. The plan and the repository do not.

The default plan artifact is a self-contained HTML file whose overview shows the outcome, the progress, and the before and after architecture. Markdown is the fallback when the repository requires it.

Workflow

  1. Identify the request: create, promote, execute, resume, or close.
  2. Read references/structure.md for the vocabulary, statuses, sections, ledger rules, and evidence rules.
  3. To create or promote, copy assets/plan-template.html, or assets/plan-template.md when the repository requires Markdown, and complete the required sections. Read references/html-format.md for the file and editing rules. When the plans directory has no index, copy assets/plans-index.html to index.html. After each edit to an HTML plan or index, run scripts/plans check on it. Run scripts/plans upgrade on a plan with older skill blocks.
  4. To execute or resume, follow references/execution.md: the goal block, the execution loop, the stop states, and closure.
  5. When execution reveals an architecture problem, apply references/architecture.md.
  6. Before delivery or a release, read references/delivery.md for acceptance, compatibility, and rollout evidence.
  7. Write plan prose with the technical-writing skill in developer mode.
  8. Apply references/context-budget.md when creating or revising resume instructions, or when repeated history crowds current work.

Core rules

  • One plan owns one outcome. Name the owner for everything out of scope.
  • Give every task verifiable success criteria: named tests, exact commands, or a measurable observation.
  • Keep one status token per ledger cell, with the proof in the evidence cell.
  • Prioritize parallelism. Run independent tasks at the same time, each in_progress with a named owner and its own worktree. A task waits only on a real dependency.
  • The orchestrator reviews delegated work before merge. It keeps the ledger and Current resume state current. It deletes merged local and remote branches and worktrees. It coordinates with peer orchestrators and never drives their plans.
  • When the commit policy permits, record each ledger transition in a plan commit right after its work commit.
  • Record evidence with exact counts and names. Mark checks that could not run UNVERIFIED instead of counting them green.
  • Keep the plan a thin control plane. Move narrative evidence to the proof root. Keep one current resume state, separate from execution history.
  • Repair the pockets of complexity that execution reveals at their owning boundaries, and route each repair through the ledger.
  • End every plan with a cleanup task that triggers when its final pull request merges. Delete the plan by default, or archive it when the repository keeps an archive.

Autonomous execution

A plan runs autonomously when its status is active and its goal block is complete. Drive it to completion through the execution loop, and stop only at a valid stop state defined in references/execution.md.

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.