Skip to content

Adonis0123/adonis-skills / skills/web-performance-audit

web-performance-audit

Verified

Run a read-only, evidence-led runtime performance audit of a real web application, especially authenticated editors, dashboards, media tools, long-lived SPAs, and pages with many tabs or states. Use when the user asks to profile slowness, inspect every tab, use Chrome DevTools or Chrome Recorder, run React Scan, compare normal and throttled interactions, find memory or rendering risks, or produce a performance report and recommendations without changing code. Discover available browser capabilities, guide the user through official installation when an important capability is missing, route to validated fallbacks, isolate measurement pollution, and distinguish measured facts from screen observations, hypotheses, and unverified paths. Do not use for a Lighthouse-only SEO/accessibility audit, tool installation with no audit goal, or when the user already wants a specific optimization implemented.

references/cross-agent-compatibility.md

View on GitHub
# Cross-agent compatibility contract

This skill uses the Agent Skills file format and keeps provider-specific behavior at the capability boundary.

## Portable package rules

- The canonical package is one directory containing `SKILL.md`, optional `scripts/`, `references/`, `assets/`, and product metadata under `agents/`.
- Keep the public identity equal to the directory/frontmatter name: `web-performance-audit`.
- Use only portable frontmatter fields required by the repository. Do not put provider-specific tool names in `allowed-tools`.
- `{{skill_path}}` is this repository's documentation placeholder for the directory containing the loaded `SKILL.md`. A host must resolve and replace it before file access; never send the literal braces to a shell. If the host cannot expose the loaded skill root, use the known installed skill directory from discovery instead of guessing from the working directory or a user's home path.
- Scripts use Python's standard library and accept explicit input/output paths.
- Instructions name capabilities first. A host may expose Chrome DevTools through MCP, a CLI, a plugin, or no direct integration.

## Discovery locations

Install from the public Adonis Skills repository:

```text
npx skills add adonis0123/adonis-skills --skill web-performance-audit
```

Otherwise, copy the complete `web-performance-audit` directory into a skill location documented by the current client. Do not copy only `SKILL.md`; the validator, references, demo assets, and metadata are part of the package. Do not guess a home directory or modify global client configuration when a project-scoped installation is sufficient.

Cursor's official Agent Skills documentation lists project-level `.agents/skills/` and `.cursor/skills/` among its discovery roots, so the same project-scoped installation works without a Cursor-only copy.

Invoke `/web-performance-audit` in Claude Code, Cursor CLI, and Grok CLI, and `$web-performance-audit` in Codex CLI. These are discovery hints, not compatibility proof.

Claude Code, Codex CLI, Cursor CLI, and Grok CLI behavior can drift by version; verify discovery with a real invocation rather than treating an installed directory as proof.

Installing this skill and installing browser capabilities are separate operations. When Chrome DevTools MCP, a structured browser, Computer Use, Recorder access, or React localization is missing, follow `{{skill_path}}/references/capability-bootstrap.md`. Never bundle a machine-specific Chrome executable, browser profile, debugging port, credential store, or user-level path into the public package.

The latest repository acceptance snapshot is `../evals/host-acceptance-2026-08-30.md`. Treat it as dated evidence: it records current passes and client-side blockers instead of converting package compatibility into a runtime guarantee.

## Host acceptance test

For each supported client, use a temporary repository with the skill installed and ask the client, in read-only/plan mode when available, to process the same sanitized fixture. A pass requires the output to:

1. identify the skill by its exact name;
2. honor the read-only boundary;
3. route unavailable tools to fallbacks instead of fabricating results;
4. preserve evidence labels and measurement caveats;
5. include coverage, recommendations, negative results, `UNVERIFIED` paths, and restoration;
6. avoid editing fixture/product files.
7. resolve bundled resources without emitting or executing the literal text `{{skill_path}}`.
8. preserve the fixture's exact target and surface names; matching only the number of tabs is a failure.
9. when a required tool is absent, give a capability-level fallback or official installation card without silently mutating the host.

Test both explicit invocation and a natural-language trigger if the host supports automatic skill selection. Record the CLI version, exit code, output, and worktree diff. “The model mentioned the skill” is not enough; verify the required behavior in the output. Grade semantic requirements rather than an unrequested JSON container type or hierarchy delimiter.

### Cursor CLI headless route

Cursor CLI supports both `plan` and `ask` as read-only modes. For headless acceptance that requires a final machine-readable artifact, use `ask` and verify the parsed result plus a clean worktree:

```text
cursor-agent --print --output-format json --mode ask --sandbox enabled \
  --workspace <temporary-repository> '/web-performance-audit <sanitized request>'
```

Only pass `--trust` for an inspected isolated repository. On the accepted 2026.08.25 CLI, the equivalent `plan` command could exit successfully after returning progress text without the requested final JSON; therefore exit code alone is not acceptance proof. Re-test this limitation before removing it.

Official Cursor references:

- https://cursor.com/docs/skills
- https://cursor.com/docs/cli/using
- https://cursor.com/docs/cli/reference/parameters

## Tool adapter rules

| Host capability         | Workflow role                                    | If unavailable                                                                |
| ----------------------- | ------------------------------------------------ | ----------------------------------------------------------------------------- |
| Chrome DevTools MCP/CLI | trace, Network, DOM, Coverage, Memory, emulation | browser probes, exported trace, visual evidence, source hypotheses; mark gaps |
| Browser plugin/driver   | exact authenticated interaction and snapshots    | Computer Use or user-provided recording; mark structured DOM gaps             |
| Computer Use            | Recorder/DevTools UI and visual-only controls    | structured driver or user-performed UI step; do not invent visual state       |
| Shell/Python            | evidence-ledger validation and rendering         | follow the report contract manually                                           |
| React Scan/Profiler     | React-localized hypothesis                       | browser trace plus source inspection; mark React ownership unverified         |

The workflow remains useful with partial capabilities because every missing layer becomes explicit coverage. It is not acceptable to silently downgrade a quantitative claim to a visual guess.

## Freshness rule

Before publishing claims that a client supports this skill, rerun the acceptance test on the locally installed client version. MCP configuration, skill discovery paths, permission flags, and non-interactive modes are version-sensitive.
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall