agentleFS
anandartinanandartin/celestial-skills/skills/incremental-implementation-engineering/AGENTS.md

celestial-skills / incremental-implementation-engineering

AGENTS.md1 starsChanged 18 days agoApache-2.0View on GitHub

Build in thin vertical slices that each compile, pass tests and get committed, so you always have a working save point and any mistake costs one small revert. Generated from the Celestial engineering practice “Incremental implementation”.

The file

We split this AGENTS.md into the rules it gives an agent

2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    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:

    Never have more than one failing test or broken thing at a time.

    line 25
  2. Distinctive:

    Commit after every green slice, and never commit red.

    line 26

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md 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
AGENTS.md

AGENTS.md — Incremental implementation

Build in thin vertical slices that each compile, pass tests and get committed, so you always have a working save point and any mistake costs one small revert.

How to do it

  1. Pick one slice — Take the next task from the plan and define the smallest piece of behaviour that works end to end. Write down the check that will prove it: usually one test.
  2. Start from green — Run the build and tests before changing anything. If they already fail, stop and report or fix that first; never start a slice on a red baseline.
  3. Make one thing fail — Add or change a single test for the slice and watch it fail for the expected reason. Keep only one failing thing at a time.
  4. Make it pass with a small diff — Write the least code that turns that test green without breaking others. Defer cleanup and generalisation to a later step.
  5. Run the full checks — Run lint, type checks and the relevant test suite. The slice is not done until all of them pass.
  6. Commit a checkpoint — Commit the green slice with a message naming the task or requirement ID. Each commit is a save point you can return to. Hide unfinished user-facing behaviour behind a feature flag so the commit could ship.
  7. Tidy, then commit again — Refactor with tests green, as a separate commit. Never mix behaviour changes and cleanup in one commit.
  8. Stop and report when stuck — If a slice is still red after a few focused attempts or a set time box, revert to the last checkpoint and report: the goal, what was tried, the exact failure output and a proposed next step. Do not pile on further changes.

Checklist (before a pull request)

  • The baseline was green before the first change
  • Every commit compiles and passes lint, types and tests
  • No commit contains more than one slice of behaviour
  • Behaviour changes and refactors are in separate commits
  • Unfinished user-facing behaviour is behind a flag that defaults off
  • Any slice that could not be made green was reverted and reported
  • The pull request diff is small enough to review in one sitting

Rules

  • Never have more than one failing test or broken thing at a time. (With one failure, the cause is the last change; with several, you are debugging combinations.)
  • Commit after every green slice, and never commit red. (Green checkpoints make rollback a single revert and keep bisecting useful.)
  • Roll back by reverting the slice's commit, not by hand-editing code back. (A revert is exact and recorded; manual undo leaves fragments behind.)
  • Build the error and empty states in their own slices, each with a test. (Edge cases added at the end tend to be skipped under deadline.)
  • Ship schema changes as additive slices before the code that uses them, and remove old columns in a later slice. (Each step stays deployable and reversible on its own.)

Checks

GateToolThresholdWhen
Green commitsCI run on every pushed commitEvery commit in the pull request passes lint, types and testsEvery push
Small diffsPull request size checkUnder 400 changed lines; no single commit over 200Every pull request

Generated from the Celestial engineering practice “Incremental implementation”.

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.