agentleFS
Sign inStart free
IgorGanapolskyIgorGanapolsky/ThumbGate/.claude/skills/scope-discipline/SKILL.md

scope-discipline

Skill29 starsChanged 7 months agoMITView on GitHub

Keep one task closed-looped before starting the next. Triggered when a new problem surfaces mid-task (a bug, a security finding, a missing env var), when you're about to branch into unrelated work, or when you notice the session has sprawled across many half-finished threads. Prevents the "nothing ever lands" failure where each step spawns new work and none finishes.

We split this skill into the rules it gives an agent

5 rules. 0 are common, 4 are distinctive, and 1 is 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.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When a new issue surfaces mid-task, do NOT start it.

    line 26
  2. Distinctive:

    After closing, pick the next item from the parking lot or ask the CEO which to take if priority is ambiguous.

    line 30
  3. Very specific:

    Keep a ## Parking lot section in the task-state-ledger:

    line 35
  4. Distinctive:

    Before branching, ask: "does this finish the current task or start a new one?" If it starts a new one → park it.

    line 49
  5. Distinctive:

    If it starts a new one → park it.

    line 50

“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

Scope Discipline — Close Loops, Don't Open Them

The failure this prevents

A single request balloons into a chain where every step uncovers new work and you chase it immediately, so nothing reaches "done + verified + deployed." This session: RAG → OAuth → directory submission → reviewer key → consent-key validation → CodeQL fixes — six threads, each started before the prior landed. The result feels like motion but lands nothing. (2026 reliability research flags "repetitive actions / no task completion" as the top long-horizon failure trigger — a metacognitive check on this is recommended: arxiv 2509.19783.)

The rule: one in-flight task, with a parking lot

  1. One task is in_progress at a time (mirror it in the task-state-ledger).
  2. When a new issue surfaces mid-task, do NOT start it. Write it to a parking lot and keep going. Park, don't pivot.
  3. A task is closeable only when verified (the evidence-first-answer / deployment-gate bar), not when code is written.
  4. After closing, pick the next item from the parking lot or ask the CEO which to take if priority is ambiguous.

The parking lot

Keep a ## Parking lot section in the task-state-ledger:

## Parking lot (found mid-task — do NOT start until current loop closes)
- [security] CodeQL: 2 critical cmd-injection + 1 XSS on main — own PR
- [infra] THUMBGATE_REVIEWER_KEY not set on Railway
- [docs] /docs/connectors 404, favicon 404

For genuinely independent work, prefer spawning a separate task (e.g. the spawn_task chip) over context-switching this session.

When the user says "continue" / "do everything"

That means make progress on the current loop, not fan out into new threads. The most valuable thing is the next closed loop, not the most new activity. Before branching, ask: "does this finish the current task or start a new one?" If it starts a new one → park it.

Allowed exception

A discovery that makes the current task unsafe to finish (e.g. you find the thing you're about to ship has a security hole) DOES belong in the current loop — fixing it IS closing the loop correctly. The test: "is this a prerequisite of the current Goal, or a new Goal?" Prerequisite → handle now. New Goal → park.

Anti-patterns

PatternFix
Find a bug mid-task → immediately go fix itPark it; finish current loop first
5 open PRs, 0 merged-and-verifiedDrive ONE to deployed before opening more
"While I'm here I'll also…"That's a new Goal — park it
Treating "continue" as "find new work""continue" = advance the current loop

Exit criteria

The current task's Goal is verified done. The parking lot is either empty or explicitly handed to the CEO as "here's what I parked; which next?"

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.

Check your setup, free