diagnostics
Skill100 starsChanged 36 days agoMITView on GitHub
Error handling and build-time diagnostics conventions - Result-not-panic, structured Diagnostics with stable codes, actionable help, color/JSON presentation layering, exit codes, and cold-path performance.
We split this skill into the rules it gives an agent
6 rules. 0 are common, 0 are distinctive, and 6 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
0
Distinctive
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 8
Use this skill whenever you add, change, or review an error path: a build-time authoring error, a parser/handler failure, a CLI validation …
- Very specific: line 48
Add dedicated variants instead of overloading a generic Generic(String) / Validation(String) so callers can match programmatically.
- Very specific: line 53
Every "the developer wrote invalid template syntax" mistake is returned as ParserError::Template(Box<Diagnostic>) …
- Very specific: line 67
Add a new authoring error with the parser helpers (authoring_error, authoring_error_at, html_error) and a new constant in diagnostic::codes.
- Very specific: line 95
When one value feeds both a terminal and a non-terminal channel, split it: webui-dev-server's RebuildError { display, message } carries a …
- Very specific: line 129
Keep hot fast-paths inlinable: a per-element check (e.g.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Error Handling & Diagnostics
Use this skill whenever you add, change, or review an error path: a build-time authoring error, a parser/handler failure, a CLI validation error, or anything surfaced to a host (FFI/WASM/Node) or a tool/agent. WebUI errors must be recoverable, actionable, and machine-consumable — for humans and AI agents alike.
1 - Never panic on recoverable input
panic = "abort" in the release profile means a panic kills the process
instantly — including any FFI/WASM/Node host embedding the framework. Bad
template input, bad CLI input, and bad state are recoverable and must return
Result, never panic!/unwrap()/expect().
| Situation | Do |
| Malformed template / CSS / route authored by a developer | Return Result with a structured Diagnostic (see §3). |
| Missing/invalid CLI input (file, port, flag) | Return a typed CliError (see §6). |
| A genuinely impossible internal state | Prefer ? with a typed error; only use unreachable!/expect with a justification comment, and never in a hot or host-reachable path. |
unwrap()/expect() are banned in library code (clippy.toml
disallowed-methods). todo!, unimplemented!, and dbg! are banned
workspace-wide (clippy.toml disallowed-macros). Tests opt out with
#[allow(clippy::disallowed_methods)].
Enforcement note.
unwrap/expect/todo!/unimplemented!/dbg!are caught by clippy.panic!is not lint-banned (too entrenched) — keep it out of recoverable paths by review. The "no regex in core logic" rule is also review-enforced, notdeny-banned:actix-webpullsregextransitively, so a crate-level ban would break the build and can't scope to first-party code.
2 - Error type conventions
| Crate kind | Error type |
Library (webui-parser, webui-handler, webui-expressions, webui-state, webui-protocol, webui-ffi) | Custom enum via thiserror. |
Binary (webui-cli, xtask) | anyhow for orchestration; a typed enum when callers must branch on the cause (e.g. webui-cli's CliError for hints + exit codes). |
- Each error layer's
Displaydescribes only its own level; the#[source]chain carries the rest, soanyhow's{:#}never double-prints. Provide a flatchain_message()helper for hosts that don't walk the chain (Node, FFI). - Add dedicated variants instead of overloading a generic
Generic(String)/Validation(String)so callers canmatchprogrammatically.
3 - Authoring errors are structured Diagnostics
Every "the developer wrote invalid template syntax" mistake is returned as
ParserError::Template(Box<Diagnostic>) (crates/webui-parser/src/diagnostic.rs),
so all build errors render identically. A Diagnostic carries:
code— a stable, machine-readable identifier (e.g.invalid-for-each). Defined indiagnostic::codes. Treat codes as a stable API: tools and AI agents branch on them, so rename only with a deliberate migration.- title — short, lowercase (
invalid <for> each expression). - location — set from a byte offset via
.at_offset(source, offset); rendered rustc-style--> owner:line:column(single forward scan, no regex, no recursion), falling back toin component <c> · element <e>. - snippet — the offending source text.
help:— an actionable fix (see §4).
Add a new authoring error with the parser helpers (authoring_error,
authoring_error_at, html_error) and a new constant in diagnostic::codes.
Validate at parse/build time and fail fast — never defer to render time.
4 - Make errors actionable (and typo-aware)
Tell the developer what is wrong and how to fix it. Every Diagnostic
should carry a help: line. Where a mistake is likely a typo, suggest the
intended name via suggest::closest_match (iterative Levenshtein — no
recursion, no regex, cold path only):
- Misspelled directive attribute:
<for eahc=…>-> "did you meaneach?" - Unknown component tag: suggest the closest same-namespace registered
component (
<mp-buton>-><mp-button>). Prefix-guard the match (text before the first-must match) so a genuine third-party custom element (<md-button>) is never falsely flagged.
5 - Presentation layering: color belongs ONLY in the entry point
Libraries produce plain, color-free data. The entry point decides how to present it. Never embed ANSI in library output or in any machine/host channel.
| Consumer | Gets |
webui-cli (terminal) | Reads Diagnostic fields and colorizes with console::style() — the only approved styling method (see copilot-instructions "Terminal output styling"). |
| FFI / WASM / Node | The plain Display text through their native error channel (webui_last_error, JsValue, napi::Error). |
Browser / tools (dev-server live-reload, SSE, console.error) | Plain text. ANSI renders as garbage and breaks single-line SSE frames. |
When one value feeds both a terminal and a non-terminal channel, split it:
webui-dev-server's RebuildError { display, message } carries a colorized
display for the reporter and a plain message for the browser.
Per-line color: when colorizing multi-line output, style each line
independently (open + close the SGR span within the line). A single span that
straddles newlines bleeds when the line is later re-prefixed (e.g. [server]
under xtask dev).
6 - Machine-readable output and exit codes (webui-cli)
For editors, CI, and AI/agent tooling:
--format json(global flag) emits each error as one JSON object on stdout (no ANSI; decorative output suppressed):{severity, code, message, file, line, column, snippet, help, chain}. Branch on the stablecode, not the humanmessage. Build the object with theserde_json::MapAPI, not thejson!macro (itunwraps internally and tripsdisallowed_methods).- Exit codes follow BSD
sysexits.h(webui-cli'serror::exit_code):65data/authoring error,66missing input,69port in use,74I/O,2usage (clap),1otherwise. - Replace fragile
err_msg.contains("...")dispatch with typed errors that own theirhint()andexit_code().
7 - Error construction is COLD - keep it off the hot path
Building a Diagnostic (format strings, suggestions, location scans) is rare,
but if it inlines into a hot function it bloats that function and perturbs its
code layout — a real, measurable regression (observed ~4-5% on parse benches
with no added hot-path work).
- Mark error builders
#[cold]+#[inline(never)](e.g.authoring_error*,html_error,css_diagnostic,*_errorconstructors,suggest::closest_match). - Keep hot fast-paths inlinable: a per-element check (e.g.
split_once('-')) must stay inlined; only its cold fallback (the registry scan) goes out-of-line. - Validate layout-sensitive changes with
cargo bench -p <crate>against the base branch (seeskills/perf/SKILL.md). A "regression" with no added compute is usually layout — fix it with#[cold], don't shrug it off.
8 - Checklist for a new error
- Returned as
Result, never panicked (host-safe underpanic = "abort"). - Authoring mistake ->
ParserError::Template(Box<Diagnostic>)with a newdiagnostic::codesconstant, location, snippet, andhelp:. help:is actionable; add a "did you mean …?" suggestion if it's a typo.- No color/ANSI in library, host, browser, or JSON output.
- Surfaced in
--format jsonwith the stablecode; exit code classified. - Error construction is
#[cold]/#[inline(never)]; hot path unchanged (confirm with a benchmark if it sits near a hot loop). - Tests: a regression test that fails without the error, asserting on the
code(not the prose); JSON stays plain (no\x1b). DESIGN.mdupdated if the error-handling architecture changed; public docs (incl.docs/ai.md) updated if a user-visible code/flag changed.
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- carbon-language/carbon-lang★ 34ksame name
- databricks/app-templates★ 217same name
- WindcleaverDev/regkit★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository11 other files it gives its agents.
- .github/copilot-instructions.mdCopilot instructions
- webui-referenceSkill
- bump-versionSkill
- code-reviewSkill
- create-exampleSkill
- docs-syncSkill
- perfSkill
- prSkill
- quality-gateSkill
- testingSkill
- webui-devSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
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.

