slicc
Copilot instructions37 starsChanged 20 days agoApache-2.0View on GitHub
Review webapp/extension/Node/Swift/iOS. Details: docs/review-patterns.md. Bound external calls; surface errors. Drain failed preload copies. Test OPFS reloads; reject live wipes. Check Node ↔ Swift endpoints/signing and browser ↔ extension mounts/VFS/secrets. FS adapters must honor encodings; test Unicode and binary pipelines without byte guessing. Preserve state across DOM rebuilds. Sprinkle activation places its parked container; serialize open/reload per sprinkle so watcher and command renders cannot overlap. Foreground the page (Page.bringToFront() / wake renderer) before screenshots or capture. Validate CDP target + port…
Where files are headingNew Copilot instructions files have more rules about branches and pull requests, and fewer about code style.
Of the Copilot instructions files we read, the 2,397 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 repository9 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- docs/CLAUDE.mdCLAUDE.md
- adding-slicc-featuresSkill
- cdp-smoke-testSkill
- demo-recordingSkill
- deploying-tray-workerSkill
- slicc-handoffSkill
- verifying-before-pushSkill
- writing-slicc-testsSkill
The file
We split this copilot-instructions.md into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
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 11
Check Node ↔ Swift endpoints/signing and browser ↔ extension mounts/VFS/secrets.
- Distinctive: line 44
Test FS limits via real shells, including custom commands.
≈ “Common” is an estimate: at least 1 in 5 of the 3,854 Copilot instructions 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
SLICC — Code Review
Review webapp/extension/Node/Swift/iOS. Details: docs/review-patterns.md.
1. Error-path coverage
Bound external calls; surface errors. Drain failed preload copies. Test OPFS reloads; reject live wipes.
2. Cross-runtime parity
Check Node ↔ Swift endpoints/signing and browser ↔ extension mounts/VFS/secrets. FS adapters must honor encodings; test Unicode and binary pipelines without byte guessing.
3. UI state preservation
Preserve state across DOM rebuilds. Sprinkle activation places its parked container; serialize open/reload per sprinkle so watcher and command renders cannot overlap.
4. CDP / Chrome edge cases
Foreground the page (Page.bringToFront() / wake renderer) before screenshots or capture.
Validate CDP target + port before trusting them; handle disconnects.
5. Native / macOS permissions
Protected access needs entitlements/usage descriptions, TCC checks, and graceful denial.
File Provider appexes must embed+sign every @rpath framework and declare network
entitlements. Avoid keychain-access-groups without an appex-specific Developer ID profile.
6. Model metadata / provider pipeline
Adobe/Copilot Anthropic proxies: clear native mid-convo flags on rebuild.
Wire messages omit output_config; retain request effort/tools.
Pi system messages hold tools: preserve both on reload/compact/clear; never summarize.
Track reasoning/input/cost/thinking through discovery→API. OpenRouter (Free): all pricing dims zero;
stream refuses IDs not in the live free catalog.
Scoops freeze model at creation; Gelatiere follows
modelFor(leadingRootOf(roster)) on boot/run/leader changes, never global selection.
7. Tests
Test FS limits via real shells, including custom commands.
Require mirrored tests and floors. Check OPFS retries, concurrent append, scoped identity, metadata errors, and ACL/sudo gates.
8. Follower surface wiring parity
Leader broadcasts need follower handlers and UI actions. Check live, follower, and extension boot paths; preserve shared fallbacks; prefer capability checks to float names.
9. Origin / bridge routing contract
Thin-bridge UI/API origins differ. Flag same-origin /api/, hardcoded origins, and
unnormalized comparisons. Iframe/channel/relay messages and UI activation must retain
the owner captured at opening, not later focus. Extension listeners:
sender.origin must be own chrome-extension://<id>.
10. Layer import direction
CI: lint:layer-back-edges; never grow baselines. Webapp fs/base → shell/git → cdp → tools → core → scoops → ui. Flag up-stack, scoops→kernel, fs/base→unranked values, cross-package imports, and git auth→shell commands (same-rung blind spot). webcomponents→webapp: none. Swift: SPM + public; no widget WebRTC. Probes below ui/ use CapabilityBroker.
11. Untyped string-keyed bags
Flag new Record<string, unknown> in source. Require a named type,
boundary validation, or a justified suppression; never grow the frozen baseline.
Cone/scoop are roles over one WorkUnit; records have no role field. Route on policy /
isRootUnit / getWorkUnits(). Tray role = ScoopSummary.parentId (#2358); .isCone is a type error.
12. Agent skill freshness
Capability/command/argument/workflow changes must update matching runtime + developer
SKILL.md files. Run skill-router + sync checks.
13. Transcript export
Require fail-closed redaction, reasoningExcluded: true, sudo approval
(kind: 'export'; only NOPASSWD Export skips it),
binary integrity, transfer-corrupt for unknown errors or SHA-256 mismatches.
Bench recovery must abort at timeout/cost cap and confirm flat spend before scoring.
14. --help that does the thing
Dispatchers checking only args[0] === '--help' route cmd <verb> --help into the
handler, which may act. Check help first, scanning all args.
Severity
🔴 Critical = likely prod issue · 🟡 Major = scenario-specific · 🔵 Minor = quality.
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.

