agentleFS
ai-ecoverseai-ecoverse/slicc/CLAUDE.md

slicc

CLAUDE.md37 starsChanged 20 days agoApache-2.0View on GitHub

Repo navigation hub. Keep package details in the nearest package CLAUDE.md and fast-changing how-to in docs/. Purposes per package: docs/root-details.md. Per-package internals: each packages/<name>/CLAUDE.md. Other top-level: docs/ (long-form docs + screenshots), packages/*/tests/ (Vitest suites), dist/ (generated; do not hand-edit). Runtime-specific guides: webapp · cherry · cloudflare-worker · node-server · cloud-core · shared-ts · webcomponents · spoon · vfs-root · go-optel · swift-launcher · swift-optel · swift-server · swift-traysession · swift-trayfollower · swift-traykit · swift-widgetkit · ios-app · slicc-cli · github-workflow…

  • Installs packages

The file

We split this CLAUDE.md into the rules it gives an agent

11 rules. About 0 are common, 2 are distinctive, and 9 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 9

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    Use ice cream terms in review comments and docs when they match the domain (e.g., feed_scoop not delegate_to_scoop).

    line 41
  2. Distinctive:

    Keep commits focused and package-local when possible.

    line 45
  3. Very specific:

    Do not hand-edit generated output in dist/.

    line 47
  4. Very specific:

    Model ID aliases: use pi-ai aliases (e.g., claude-opus-4-6), not dated snapshot IDs.

    line 77
  5. Distinctive:

    Every change must satisfy tests, docs, and verification.

    line 82
  6. Very specific:

    Add/update tests for behavior changes.

    line 86
  7. Very specific:

    Run the full pre-push/PR pass — lint (first.

    line 101
  8. Very specific:

    Add/change a SLICC feature surface → use adding-slicc-features

    line 107
  9. Very specific:

    Write/update SLICC tests → use writing-slicc-tests

    line 112
  10. Very specific:

    Commit, push, open/update a PR, diagnose verification CI failures → use verifying-before-push

    line 113
  11. Very specific:

    When you change a category, update docs/review-patterns.md (source of truth) and the ≤4,000-char .github/copilot-instructions.md so all …

    line 134

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

CLAUDE.md

Repo navigation hub. Keep package details in the nearest package CLAUDE.md and fast-changing how-to in docs/.

Module Map

Purposes per package: docs/root-details.md. Per-package internals: each packages/<name>/CLAUDE.md.

  • Web / TS: packages/webapp/, packages/webcomponents/, packages/cherry/, packages/spoon/ (the Chrome extension lives in ai-ecoverse/slicc-extension)
  • Node / cloud: packages/node-server/, packages/cloudflare-worker/, packages/cloud-core/, packages/shared-ts/, packages/vfs-root/, packages/dev-tools/, packages/assets/, packages/github-workflow/, packages/bench/
  • Native (Swift / iOS / Go): packages/swift-launcher/, packages/swift-server/, packages/swift-optel/, packages/swift-traysession/, packages/swift-trayfollower/, packages/swift-traykit/, packages/swift-widgetkit/, packages/ios-app/, packages/slicc-cli/, packages/go-optel/

Other top-level: docs/ (long-form docs + screenshots), packages/*/tests/ (Vitest suites), dist/ (generated; do not hand-edit).

Top-Level Commands

npm install                              # Install dependencies (first time)
npm run build                            # Production build (all workspaces)
npm run build -w @slicc/webapp           # UI-only build (faster for UI changes)
npm run test                             # Vitest run
npm run typecheck                        # Browser + Node typecheck
npm run dev                              # Thin /cdp bridge + Chrome (UI from hosted origin)

Runtime-specific guides: webapp · cherry · cloudflare-worker · node-server · cloud-core · shared-ts · webcomponents · spoon · vfs-root · go-optel · swift-launcher · swift-optel · swift-server · swift-traysession · swift-trayfollower · swift-traykit · swift-widgetkit · ios-app · slicc-cli · github-workflow · bench · dev-tools · docs

External Handoffs

handoff to slicc / move this to slicc opens https://www.sliccy.ai/handoff?handoff=<text> (or ?upskill=<url>) with a handoff:/upskill: verb-prefixed instruction. Worker returns an RFC 8288 Link header SLICC observes on navigation for approval. Prefer .agents/skills/slicc-handoff/scripts/slicc-handoff.

Ice Cream Vocabulary

  • Cone: main agent. Full filesystem access, all tools.
  • Scoops: isolated sub-agents; sandboxed filesystems (/scoops/{name}/ + /shared/). Tools: scoop_scoop, feed_scoop, drop_scoop.
  • Licks: external events — webhooks, cron, workflow completions.
  • Floats: runtime envs — CLI, extension, Electron, cloud (hosted-leader), Cherry (embedded follower garnish; ?cherry=1 in a host page's iframe).
  • Biscotto (pl. biscotti): a revocable guest seat on a cone. Holder gets a private *.sliccy.now URL showing the live transcript; its messages are untrusted by default and pass a message/tool gate. Restricted at the wire by scoops/tray-leader/biscotto-gate.ts. Approvers: user, cone, agent (a bounded per-request approver driven by /etc/APPROVALS.md), or a delegated scoop.

Use ice cream terms in review comments and docs when they match the domain (e.g., feed_scoop not delegate_to_scoop).

Git Conventions

  • Keep commits focused and package-local when possible.
  • Linear history: CI rejects merge commits. Rebase onto origin/main; Husky enforces via packages/dev-tools/tools/check-linear-history.sh.
  • Do not hand-edit generated output in dist/.
  • Auth: GitHub OAuth login (preferred; renews the git bridge on network ops) or git config github.token <PAT>; see docs/secrets.md.

Requires Node >= 22 (LTS). Ports: 5710 (bridge + /api), 9222 (Chrome CDP), 9223 (Electron CDP). node-server serves no UI — the webapp loads from the hosted origin and dials back to the local /cdp bridge.

Parallel Instances

Override the UI port to run parallel standalone instances; profiles and CDP ports stay isolated:

PORT=5720 npm run dev   # Second instance on port 5720
PORT=5730 npm run dev   # Third instance on port 5730

Principles

  1. Claw pattern — persistent browser orchestration layer over Pi.
  2. Virtual CLIs first — composable shell commands over dedicated tools.
  3. Browser-first — state client-side; the server is a stateless relay; the extension has zero server.
  4. Skills over hardcoded features — add capabilities as SKILL.md files when possible.

Architecture

SLICC runs as standalone CLI (Express + Chrome), Chrome extension, Electron float, cloud hosted leader, or Cherry follower. See docs/architecture.md + float topology. Each package CLAUDE.md is authoritative for its subsystem. Shell: docs/shell-reference.md. Verification/CI gates: .agents/skills/verifying-before-push/SKILL.md.

Key Conventions

  • Tests: packages/*/tests/ mirrors src/. Vitest, globals: true, environment: node. fake-indexeddb/auto for VFS tests.
  • Dual-mode compatibility: features MUST work in CLI and extension. The thin extension runs no dynamic code — UI, sprinkles/dips, JS realms, WASM all run in the hosted leader tab / kernel worker; extension assets load via chrome.runtime.getURL(). The extension itself is in ai-ecoverse/slicc-extension: an extension-protocol change in webapp (kernel/messages.ts, offscreen/picker clients) or shared-ts must also land there.
  • Extension detection: isChromeExtensionRealm() from @slicc/shared-ts (lint-gated).
  • Model ID aliases: use pi-ai aliases (e.g., claude-opus-4-6), not dated snapshot IDs.
  • Developer vs agent CLAUDE.md: developer-facing CLAUDE.md lives at repo root and each package. Agent-facing runtime CLAUDE.md is packages/vfs-root/shared/CLAUDE.md (bundled as /shared/CLAUDE.md). Tier table: docs/CLAUDE.md.

Change Requirements

Every change must satisfy tests, docs, and verification.

Tests

  • Add/update tests for behavior changes; TS tests in packages/*/tests/, mirrored by subsystem. Patterns: .agents/skills/writing-slicc-tests/SKILL.md.
  • Coverage thresholds enforced in CI. Floors in coverage-thresholds.json; raised by the nightly ratchet (packages/dev-tools/tools/coverage-ratchet.mjs). Never hand-lower them. TS: npm run test:coverage:<package>; Swift: packages/dev-tools/tools/swift-coverage-check.sh.

Documentation

Doc tier map (full table: docs/CLAUDE.md):

  • README.md — user-facing behavior changes
  • CLAUDE.md files — developer conventions, architecture, builds
  • docs/ — detailed tools, commands, patterns
  • vfs-root/workspace/skills/*/SKILL.md — agent skills (shell command changes)
  • .agents/skills/*/SKILL.md — developer agent skills (verification, wiring, tests, runbooks)

Verification

Run the full pre-push/PR pass — lint (first; top CI failure), typecheck, test, test:coverage, both builds, plus the touched-file debt gate — before committing. Commands and CI-only gates: .agents/skills/verifying-before-push/SKILL.md; CI in .github/workflows/ci.yml.

Developer Agent Skills (.agents/skills/)

Skill-aware harnesses load these directly; this list routes AGENTS.md-only harnesses to them.

  • Add/change a SLICC feature surface → use adding-slicc-features
  • Deploy/debug the Cloudflare tray hub worker → use deploying-tray-worker
  • Record a UI demo for a PR → use demo-recording
  • Hand work off to SLICC → use slicc-handoff
  • Smoke-test a build in a controlled browser → use cdp-smoke-test
  • Write/update SLICC tests → use writing-slicc-tests
  • Commit, push, open/update a PR, diagnose verification CI failures → use verifying-before-push

Automated PR Review Checklist

Reviewers (Claude action, Codex AGENTS.md, Copilot .github/copilot-instructions.md) and humans check PRs against these blind spots. Descriptions: docs/root-details.md; full catalog: docs/review-patterns.md.

  1. Error-path coverage
  2. UI state preservation
  3. Cross-runtime parity
  4. CDP edge cases
  5. Native/macOS permissions
  6. Model metadata / provider pipeline
  7. Test coverage
  8. Follower wiring parity
  9. Origin/bridge routing
  10. Layer import direction
  11. Untyped string-keyed bags (Record<string, unknown>)
  12. Agent skill freshness
  13. Transcript export
  14. --help that does the thing

When you change a category, update docs/review-patterns.md (source of truth) and the ≤4,000-char .github/copilot-instructions.md so all reviewers stay in sync.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.