token-discipline / snippets
AGENTS.md1 starsChanged 8 months agoMITView on GitHub
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository2 other files it gives its agents.
- snippets/CLAUDE.mdCLAUDE.md
- token-disciplineSkill
The file
We split this AGENTS.md into the rules it gives an agent
10 rules. About 0 are common, 10 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
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
- Distinctive: line 14
high: only for explicit research, deep debugging, architecture, audits, or high-stakes tasks
- Distinctive: line 18
Do not use tools for direct questions unless essential.
- Distinctive: line 19
Search before read, snippet before whole file.
- Distinctive: line 21
Stop after the first likely root cause for low-stakes diagnosis.
- Distinctive: line 22
Do not verify the same claim multiple ways unless stakes justify it.
- Distinctive: line 23
Do not replay full history on follow-ups.
- Distinctive: line 24
Do not rerun work for duplicate queued messages.
- Distinctive: line 26
Run smoke tests, benchmarks, evals, and other noisy probes in isolated sessions only, never in the active user-facing session.
- Distinctive: line 27
If background work is still running when a new user message arrives, answer the user first, then resume, isolate, or cancel the background …
- Distinctive: line 36
If the cheap path is inconclusive, explicitly say you are escalating and why.
≈ “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.
Files
Token Discipline
Task routing
Classify each request before acting:
- direct question
- simple execute
- diagnosis
- research
- writing / ideation
Budget modes
- low: max 2 tool calls on first pass, no browser by default, no full-file reads unless necessary, reply <=120 words unless asked for detail
- medium: max 5 tool calls on first pass, targeted reads only, browser only if cheaper checks fail
- high: only for explicit research, deep debugging, architecture, audits, or high-stakes tasks
Hard defaults
- Answer direct questions in the first sentence.
- Do not use tools for direct questions unless essential.
- Search before read, snippet before whole file.
- API/CLI before browser when possible.
- Stop after the first likely root cause for low-stakes diagnosis.
- Do not verify the same claim multiple ways unless stakes justify it.
- Do not replay full history on follow-ups. Use deltas.
- Do not rerun work for duplicate queued messages.
- Direct user messages pre-empt background smoke tests, evals, and diagnostics.
- Run smoke tests, benchmarks, evals, and other noisy probes in isolated sessions only, never in the active user-facing session.
- If background work is still running when a new user message arrives, answer the user first, then resume, isolate, or cancel the background work.
Output defaults
- direct question: 1-3 lines
- simple execute: short confirmation + result
- diagnosis: cause first, then fix/next step
- writing: 2-5 strong options
Escalation
If the cheap path is inconclusive, explicitly say you are escalating and why.
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.