agent-budget
Skill5 starsChanged 2 months agoMITView on GitHub
size a delegated task before you start it, and read the signals that say you sized it wrong
1 other project uses a version of this.
Adopted by
Most recently taken up by
- EdwardAF-IT/claude-skillstheir own version, first seen Nov 1, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- EdwardAF-IT/claude-skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository8 other files it gives its agents.
- code-reviewSkill
- commit-disciplineSkill
- dependency-checkSkill
- eval-before-mergeSkill
- incident-reviewSkill
- plan-firstSkill
- safe-refactorSkill
- test-firstSkill
The file
Files
agent-budget
every delegated task gets a budget whether or not i set one. sizing it up front:
- name the artifact it hands back. one diff, one file, one answer. a task that returns "an understanding" has no ceiling and will use all of it.
- count what it must read, not what it writes. a one-line fix in code nobody has mapped is the expensive one; the reading is the spend.
- give the first attempt the smallest budget that could plausibly work. it's a probe - its job is to tell me the real size, not to finish.
- write the stop down before it starts: how many attempts, and what happens after the last.
plan-first already asks for the files you expect to touch and the check for "done". that answer is the budget - if the plan can't name them, sizing is guesswork.
reading the miss
- too small: output truncates, it re-derives what it already established, the same file gets read three times. don't top it up mid-run - the run is the wrong shape. re-scope and restart.
- too big: it did the ask then kept going. room left over gets spent, so treat gold-plating as a sizing symptom and let eval-before-merge separate the ask from what arrived with it.
- raise once. still short means it's two tasks: split when half is survey and half is change, and hand the reading on as an artifact (patterns/multi-agent-handoff.md).
guardrails caps an autonomous run; this is that ceiling scoped to one handed-over task. a budget is a claim about how big the task is - when the run breaks it, fix the claim.
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
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.