Codex-Workspace
AGENTS.md2 starsChanged 56 days agoMITView on GitHub
These instructions apply to Codex Workspace and repos/workspace-hub unless a deeper repo-specific AGENTS.md overrides them. For first-time repo intake, use docs/09-new-repo-baseline.md. Every intake must assess Graphify eligibility as code, documents, mixed, assets, or sensitive. Do not install Graphify, generate a graph, or add hooks automatically. Adoption requires a repo-specific .graphifyignore, an explicit graphify-out/ ignore rule, and passed privacy/correctness gates. Maintain a clean, high-performance mixed-repo workspace. Never create one shared node_modules or equivalent install directory across unrelated repos. Expected top-level folders:
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 repository10 other files it gives its agents.
- .github/copilot-instructions.mdCopilot instructions
- agent-tooling-doctorSkill
- quality-gateSkill
- repo-onboardingSkill
- runtime-triageSkill
- security-checkSkill
- ssh-keychain-accessSkill
- typescript-reviewSkill
- workspace-maintenanceSkill
- workspace-quality-gateSkill
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
13 rules. About 0 are common, 10 are distinctive, and 3 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.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 10
Every intake must assess Graphify eligibility as code, documents, mixed, assets, or sensitive.
- Very specific: line 28
Never create one shared node_modules or equivalent install directory across unrelated repos.
- Distinctive: line 51
Keep agent context small by default.
- Distinctive: line 53
Do not load or summarize these unless the task explicitly needs them:
- Distinctive: line 69
Avoid broad historical docs and deep/artifact search by default.
- Distinctive: line 93
When no manifest exists, classify cautiously:
- Distinctive: line 102
Do not add background memory hooks or leave memory closeout as a manual reminder.
- Distinctive: line 109
run git status --short before closing so the handover does not imply a cleaner worktree than exists
- Very specific: line 110
if public docs changed, keep README.md, docs/README.md, docs/CHANGELOG.md, and relevant repo-local docs aligned
- Distinctive: line 125
Keep its targets named and narrow.
- Distinctive: line 137
When a user request appears to need a skill, check the available local skill sources before creating, importing, or researching a new one.
- Distinctive: line 157
When asked to update reviewed GitHub refs or managed upstream mirrors, use:
- Distinctive: line 162
Do not refresh those sources by hand when a wrapper covers the flow.
≈ “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
AGENTS.md
Scope
These instructions apply to Codex Workspace and repos/workspace-hub unless a
deeper repo-specific AGENTS.md overrides them.
For first-time repo intake, use docs/09-new-repo-baseline.md.
Every intake must assess Graphify eligibility as code, documents, mixed,
assets, or sensitive. Do not install Graphify, generate a graph, or add
hooks automatically. Adoption requires a repo-specific .graphifyignore, an
explicit graphify-out/ ignore rule, and passed privacy/correctness gates.
Priorities
Maintain a clean, high-performance mixed-repo workspace.
Prefer:
- clear structure
- independently runnable repos
- shared caches, not shared installs
- direct local dev servers for frontend/WebGL projects
- optional proxy or mapped-host tooling only when it adds value
- pragmatic WordPress handling, usually through Local for existing sites
Never create one shared node_modules or equivalent install directory across
unrelated repos.
Workspace Shape
Expected top-level folders:
docs/repos/tools/cache/shared/
Use:
docs/for canonical workspace docs and handover.workspace/project.jsonfor per-repo runtime metadatatools/templates/for starter templatesshared/for durable workspace metadatacache/for generated summaries, runtime artifacts, and shared stores
Token Budget
Keep agent context small by default.
Do not load or summarize these unless the task explicitly needs them:
docs/archive/- repo
ref/ - repo
screenshots/ cache/- generated reports, large copied HTML, archives, lockfiles, vendor/build output
For fresh context, read:
- this file
docs/HANDOVER.md- the repo-local
CONTEXT_CATALOG.mdor established catalog index when present - the relevant repo README, handover, manifest, or source files routed by that catalog
- generated
cache/context/.../entry.mdonly if useful
Avoid broad historical docs and deep/artifact search by default.
For repositories with recurring context-navigation costs, assess the optional
tracked context-catalog workflow in docs/09-new-repo-baseline.md. Preserve an
existing catalog hierarchy instead of creating a second one. A catalog routes
to authoritative files; it does not become authoritative memory itself.
Workspace Hub
For repos/workspace-hub, prefer:
- readable modular code
- explicit runtime/process handling
- conservative repo classification
- graceful status and failure messages
- existing patterns over new abstractions
Avoid:
- assuming all repos use the same package manager
- assuming proxy mode is better than direct local preview
- hard-coding unstable absolute paths beyond the workspace root
- hidden auto-installs or heavy setup steps
When no manifest exists, classify cautiously:
- Vite, static, Three.js, and WebGL repos default to
direct - WordPress projects already managed elsewhere default to
external
Workspace Memory
The previous workspace memory service has been removed.
Do not add background memory hooks or leave memory closeout as a manual reminder. Record closeout in tracked docs instead.
For handover updates:
- repo-specific updates go in repo-local docs and
docs/HANDOVER.mdwhen relevant - workspace-level updates go in
docs/HANDOVER.mdanddocs/CHANGELOG.md - run
git status --shortbefore closing so the handover does not imply a cleaner worktree than exists - if public docs changed, keep
README.md,docs/README.md,docs/CHANGELOG.md, and relevant repo-local docs aligned - for an adopted Graphify repo, finalize source and handover text, refresh the
local graph through
tools/scripts/graphify-repo.sh --run <repo-path>, and make no further source edits without refreshing again
Local-model context handover
The optional local-first worker at tools/scripts/local-model-context.sh is a
task-time preparation helper, not a background memory service. Its public
profile defaults to loopback Ollama for bounded extraction, classification,
inventory, routing suggestions and draft handovers. Optional external routes
must come from an explicitly selected private profile and may receive only
named, tracked, public-safe files. GPT and the operator remain responsible for
authority, privacy, public claims, destructive actions and final approval.
Keep its targets named and narrow. It must not auto-scan all repos, install or
download models, access protected paths by default, or write source material
into generated cache packets. Keep provider keys in macOS Keychain and out of
tracked files, logs and prompts. Use docs/24-local-model-context-handover.md
for the packet contract and privacy boundary.
Public Agent Distribution
Root AGENTS.md is the canonical public agent contract, including for
TomeVault distribution. .github/copilot-instructions.md is a standalone
Copilot summary, not the source of truth for cross-platform conversion.
When a user request appears to need a skill, check the available local skill
sources before creating, importing, or researching a new one. Start with the
current session's advertised skills, then inspect relevant workspace skill
folders such as .agents/skills/, shared/skills/, $CODEX_HOME/skills or
~/.codex/skills when needed. Local operators may also maintain ignored or
private skill libraries outside the public distribution tree. Prefer using or
adapting an existing local skill over adding another copy.
Tracked skills published for TomeVault live under .agents/skills/ and are
mirrored from tools/manifests/tomevault-skills.json by
tools/scripts/sync-tomevault-skills.sh. Do not install TomeVault Relay, add
TomeVault badges, or commit generated multi-format TomeVault files unless that
is explicitly requested.
Private skill review and archival destinations are operator configuration and must not be hard-coded into the public agent contract. Do not rediscover local review queues, generated outputs or legacy aliases as new public source skills.
Upstream Updates
When asked to update reviewed GitHub refs or managed upstream mirrors, use:
tools/scripts/manage-workspace-capabilities.shfor abilities and core servicestools/scripts/update-github-refs.shfor dry runs or applied updates
Do not refresh those sources by hand when a wrapper covers the flow.
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.

