brandonperfetti/agent-working-agreements

The working agreement between Brandon Perfetti and every AI agent he works with — Claude Code,
Cowork, Codex, OpenClaw (Camina) — across the repositories he runs under Sans Faux. It is the
written form of deciding what gets built and holding agents to it: how orchestrated work is
specified (a grilling session, a master priority document, tickets, waves of agents), how material
findings are labelled as measured, read from source or inferred, which branch a change may land on
and who merges it, and what an agent must stop and ask about rather than decide. The agreement
itself is AGENT-WORKING-AGREEMENTS.md — start there, not here.
This README is for a human reader: the maintainer, or someone arriving from outside.
## Reading this repo
These are one engineer's working agreements, running against a real fleet. The paths
(~/dev/sansfaux/agent/), the client appendices and the initiatives named in the changelog are
his, and they stay as they are: agents read the core, the studio profile and their client file
before a session's first delivery, and a path an agent can follow is worth more than one a
stranger finds tidy. What is portable is the core:
AGENT-WORKING-AGREEMENTS.md holds the evidence labels, the seats,
the two modes and the guard that picks between them, and the stop-list, with no owner, staff,
account, path or profile name in it; used without a profile, it follows the conventions the owner
and repository actually use. Brandon's studio — his identity, the git ritual, the store, the trigger
index and the client index — is bound in profiles/sansfaux.md. This is
the real thing, not a template. The changelog is the record of these rules being exercised: an
entry names what produced it, and a rule earns its place only after something bit. The
documentation is licensed CC BY 4.0 (the text is in LICENSE); the shell scripts under
scripts/ and the GitHub workflows are MIT (the text is in LICENSE-MIT), copyright 2026
Brandon Perfetti.
## Layout
text
AGENTS.md repository-specific agent instructions
.github/copilot-instructions.md a symlink to AGENTS.md, the name GitHub Copilot reads
.github/pullrequesttemplate.md the pull request template
AGENT-WORKING-AGREEMENTS.md the portable core: composition, Parts A (discipline) + B (modes)
profiles/sansfaux.md the studio profile: bindings, studio authority, ritual, indexes
procedures/ .md the mechanics behind a rule, read when its trigger occurs
procedures/ / .md the same, for one client, linked from its appendix
history/openclaw.md the shared client history archive, one section per client: never required reading
clients/ .md one appendix per environment — read only your own
docs/agents/issue-tracker.md repository-specific issue/spec fetch and PR-linkage rules
docs/agents/triage-labels.md the triage roles mapped to this repository's labels
docs/agents/domain.md the agreement's terms, and GLOSSARY.md as where resolved ones go
docs/agents/commit-subject-audit.md the 2026-09-23 audit of commit-subject conventions across active repositories
CHANGELOG.md why each rule changed, newest first
LICENSE the CC BY 4.0 legal code; it covers the documentation
LICENSE-MIT the MIT text; it covers scripts/ and the GitHub workflows