ai-floppy
AGENTS.md2 starsChanged 2 months agoMITView on GitHub
main is protected and refuses a direct push — from every session, including the one that owns this repository. If your push comes back with GH013: Repository rule violations found, nothing is wrong: branch, push the branch, open a pull request. No approval is required — a PR here can be self-merged. What the rule buys is not a second person, it is that tests.yml runs on pull_request, so both platforms are checked before the change is on main, and…
- Commits and pushes
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 31,700 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 repository7 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- agent-memorySkill
- consolidateSkill
- initSkill
- startSkill
- workstatusSkill
- wrapSkill
The file
We split this AGENTS.md into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
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
- Very specific: line 15
No approval is required — a PR here can be self-merged.
≈ “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
Changes land through pull requests
main is protected and refuses a direct push — from every session, including
the one that owns this repository. If your push comes back with GH013:
Repository rule violations found, nothing is wrong: branch, push the branch,
open a pull request.
git switch -c <topic>
git push -u origin HEAD
gh pr create --fillNo approval is required — a PR here can be self-merged. What the rule buys is
not a second person, it is that tests.yml runs on pull_request, so both
platforms are checked before the change is on main, and that a change
arriving from outside the session working on this repository is visible as a
change rather than as history.
That last case is ordinary here, not an incident. knowledge/ is a
cross-project base by design: an agent working in another repository that
learns something worth keeping is expected to contribute a note. It is also
the surface with the least automation behind it — most notes carry nothing
executable, so knowledge-recheck.py cannot see them, and knowledge-rot-check.py
measures age, which says nothing about a claim that was wrong on the day it was
written. Review is the only check those notes get.
The cross-project memory layer here is common/, not basic-memory
.claude/settings.json denies the basic-memory MCP server in this repository,
and deniedMcpServers keeps it from connecting at all. This overrides the
global rule that names that server as the cross-project memory layer, for this
repository only — the server stays connected and unchanged everywhere else.
Where a cross-project fact goes here instead:
knowledge/notes/— a verified finding about harnesses, shell or git that passes the three admission criteria inknowledge/README.md. Public, English, published to the site, reviewed through a pull request..agent-memory/common/private— cross-project but personal: a measured setting, an evaluation of an external tool, a trick that lives outside any checkout.
Why, in one line: basic-memory's store is local-only and not under git, so it
does not travel between machines — which is the one thing a cross-project layer
exists for — while common/ is wired by store/workplace and read by start.
The full argument, with what moved and when to revisit, is in the memory note
basic-memory-is-off-in-this-repository.
Which half of the memory a task belongs to
MEMORY.md is a router: two always-read notes and one link per half. start
opens the half the task belongs to and leaves the other two closed, so the
routing has to be answerable before reading any of them. It is written here
because which words point where is this repository's knowledge, not the
plugin's.
| the task is about | half | words that route to it |
notes, scopes, stores, quota.lock, the status files, another project's memory, whether any of it pays | memory/ | note, index, scope, store, common/, NOW.md, recall, benchmark |
scripts/, shim/, skills/, tests/, the generated site, knowledge/ | product/ | verb, shim, guard, test, suite, bash 3.2, site, search, lint |
branches, pull requests, the protected main, the workflows that run on them | delivery/ | branch, push, PR, merge, release, workflow, CI |
Split on 2026-09-09 from one flat index that had reached pointers_max=25.
The frozen rule is that a full index splits and the number stays, so if a half
fills up it splits again into sub-indexes — three levels is the floor of the
tree, not a budget to raise.
Two things the table cannot decide, and both go to memory/: a fact about the
memory of another project (this repository's product is that memory, so its
own memory is where the subject is studied), and a fact that is not about this
project at all — the latter belongs in common/, which no index links to and
start reaches by name.
<!-- floppy:agents-section -->Agent memory
This repository uses the floppy plugin for its session ritual and its
durable memory. Its verbs are run from the plugin, not from here:
bash <plugin>/scripts/run <verb>, where <plugin> is two directories above
the base directory the harness states when it loads a floppy skill
(Base directory for this skill: <plugin>/skills/start). See agent-memory
for what a note looks like and how the memory is laid out, and
start / workstatus / wrap for the three rites
built on top of it. Settings live in .floppy/config; the memory itself is
under .agent-memory.
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.

