slicc
CLAUDE.md37 starsChanged 20 days agoApache-2.0View on GitHub
This file covers the documentation surface in docs/. All package CLAUDE.md files are now within the 20,000-char cap. The grandfathered exemption list in check-doc-sizes-lib.mjs is empty. Exemption values may only be lowered or deleted — never added or raised. Do not add "Added in PR #NNN" or "Changed in PR #NNN" lines to CLAUDE.md files. Provenance belongs in git log; durable lessons (gotchas, anti-patterns, decisions) belong in docs/pitfalls.md or docs/review-patterns.md, not inline breadcrumbs. Deep implementation detail that does not fit…
In this repository9 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- adding-slicc-featuresSkill
- cdp-smoke-testSkill
- demo-recordingSkill
- deploying-tray-workerSkill
- slicc-handoffSkill
- verifying-before-pushSkill
- writing-slicc-testsSkill
The file
We split this CLAUDE.md into the rules it gives an agent
6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
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
- Distinctive: line 16
Update the nearest package CLAUDE.md when a change is package-specific.
- Distinctive: line 17
Update the root CLAUDE.md only for repo-wide navigation, CI gates, or cross-cutting principles.
- Very specific: line 18
Put long-form implementation detail in the appropriate docs/*.md file rather than bloating a CLAUDE.md.
- Very specific: line 29
All package CLAUDE.md files are now within the 20,000-char cap.
- Very specific: line 35
Do not add "Added in PR #NNN" or "Changed in PR #NNN" lines to CLAUDE.md files.
- Distinctive: line 89
Keep this directory explanatory, not redundant: prefer one authoritative page per topic and link to it from the shorter CLAUDE.md files.
≈ “Common” is an estimate: at least 1 in 5 of the 28,029 CLAUDE.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
CLAUDE.md
This file covers the documentation surface in docs/.
Documentation Tiers
| Tier | Primary file/location | Purpose |
| Public | README.md | User-facing overview and onboarding |
| Development | root and package CLAUDE.md files | High-signal developer guidance and package navigation |
| Agent reference | docs/ | Detailed architecture, commands, patterns, pitfalls, and workflows |
| Developer agent skills | .agents/skills/*/SKILL.md | Repo procedures for verification, feature wiring, tests, and operations runbooks |
How to Update Docs
- Update the nearest package
CLAUDE.mdwhen a change is package-specific. - Update the root
CLAUDE.mdonly for repo-wide navigation, CI gates, or cross-cutting principles. - Put long-form implementation detail in the appropriate
docs/*.mdfile rather than bloating aCLAUDE.md.
Size budgets (enforced by npm run lint:docs)
| File | Limit | Unit | Note |
CLAUDE.md (root) | 15,000 | chars | enforced |
packages/*/CLAUDE.md | 20,000 | chars | enforced |
packages/vfs-root/shared/CLAUDE.md | 2,048 | bytes | bundled into the VFS |
.github/copilot-instructions.md + *.instructions.md | 4,000 | chars | Copilot truncation limit |
All package CLAUDE.md files are now within the 20,000-char cap. The grandfathered
exemption list in check-doc-sizes-lib.mjs is empty. Exemption values may only be
lowered or deleted — never added or raised.
No PR breadcrumbs
Do not add "Added in PR #NNN" or "Changed in PR #NNN" lines to CLAUDE.md files.
Provenance belongs in git log; durable lessons (gotchas, anti-patterns, decisions)
belong in docs/pitfalls.md or docs/review-patterns.md, not inline breadcrumbs.
Deep implementation detail that does not fit the budget should move to docs/, not
shrink to a one-liner in a CLAUDE.md.
Common Destinations in docs/
Architecture and build:
architecture.md— detailed subsystem/file maps, layer stack, IndexedDB inventory, tray/sync matrix (opens witharchitecture-diagram.png, the float topology overview)development.md— build, run, and debug workflows- writing-slicc-tests — testing patterns and command selection
- verifying-before-push — pre-push/PR validation pass: lint internals, the boy-scout debt gate, coverage floors, and other CI-only gates
- adding-slicc-features — how to add a new shell command, tool, provider, sprinkle, etc.
kernel/process-model.md— kernel-host / process-manager deep referencework-unit.md— ADR for #1666: cone and scoop as roles over oneWorkUnitruntime; policy table, invariants, migration phasesgelatiere.md— the gelatiere, SLICC's resident advisor: a persistent root unit that reviews sessions nightly and after a chat ends and suggests skills and use cases; why it is a root, stores, delivery, the welcome-card streamlayouts.md— the panel system (SliccPanel+<slicc-layout>), layout documents and their two VFS roots, thelayoutshell command, locking, the Cherry-pushed-layout wire path, and the dock-tree it supersedesfeature-flags.md— typed registry, worker overlays, Cherry host keys,lint:dead-flags, and the propose → graduate → retire pathpanel-system-design.md— why the panel system is shaped the way it is: the trust model for agent-authored panels, docked-vs-floating, and the two hardening measures (approval-path capture, trusted layer)
Subsystems:
shell-reference.md— shell command reference (authoritative per-command list)computer-protocol.md—computerwire types, registry, adapters (v86/tab/jsh), screenshot-space coordsplaywright-cli-sync.md— how Slicc'splaywright-clistays aligned with the official@playwright/cli: the sync script, the Slicc command manifest, when to run, and how to add a new commandtools-reference.md— agent tool surface referencemounts.md—mountsetup for local FS Access, S3/R2/MinIO, and Adobe da.livesecrets.md— secrets storage, masking, and domain-scoped injectionapprovals.md— capability approval gates: shared pattern + authority axis, sudo policy, device & gesture gates, OS capture gatesoauth-intercept.md— provider OAuth intercept and silent renewaloperational-telemetry.md— Helix RUM beacons and debug samplingslicc-handoff.md— external handoff protocol (RFC 8288Linkheader +navigatelick).agents/skills/deploying-tray-worker/SKILL.md— tray hub ops runbook: R2 asset archive setup, deploy-gating, ghost-leader reconnect, CI retry logic, and staging test guidelink-discovery.md— the standalonediscovershell command and the--discoverflag onplaywright-clisubcommands (fetch,goto,navigate,open,tab-new); covers RFC 8288 / RFC 9727 parsing, emission, and the SLICC handoff/upskill relsurls.md— production URL inventoryelectron.md— Electron float workflowsliccstart-browser.md— Sliccstart's browser-like surfaces: the macOS default-browser role (registration, Info.plist declarations, incoming-link routing over CDP) and previous-session tab restoreios-simulator-qa.md— hand-running the iOS follower in a Simulator: boot/build/install/launch, seeding@AppStoragevia launch arguments, getting a real Join URLbench-runner.md— benchmark runner internals: judge (incl. fallback), none-skills flag, transcript collection over the 8 MiB tray limit, leader lifecycle and diagnostics, publishing pipeline
Review & gotchas:
pitfalls.md— runtime and extension gotchas (CSP, dual-mode runtime detection, WASM heap views, etc.)review-patterns.md— source of truth for the automated PR reviewers (Claude action, Codex, Copilot): the numbered blind-spot categories, the five-runtime parity matrix, and the severity rubric
Other:
exploration/— open design notes (not load-bearing reference)screenshots/— image assets used by README and other docs
Planning artifacts (docs/superpowers/specs/, docs/superpowers/plans/) are intentionally not kept on main — .gitignore keeps them out. Keep them local or in the PR description.
Keep this directory explanatory, not redundant: prefer one authoritative page per topic and link to it from the shorter CLAUDE.md files.
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.

