agentsclimarketplace

Foxcode acceptance testing

Skill korchasa/foxcode/.agents/skills/foxcode-acceptance-testing

Run FoxCode acceptance testing when asked to verify that FoxCode works end-to-end, smoke-test browser automation, validate evalInBrowser, check Claude Code/OpenCode/Codex integration, or prove a release candidate works with real Firefox.From its SKILL.md

Install
npx -y skills add korchasa/foxcode --skill foxcode-acceptance-testing

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.8 KB, as published. Nobody here has run it

FoxCode Acceptance Testing

Use this skill to test behavior, not packaging. For install/update/package contents, use foxcode-distribution-testing.

Ground rules

  • Read documents/requirements.md and documents/design.md before testing.
  • Use relative paths in commands from the repository root.
  • Do not skip a tier silently. If a prerequisite is missing, record it as blocked with the command output.
  • Do not edit code unless the user asked to fix failures. For a test-only request, stop at diagnosis.
  • Keep evidence: command, result, and the exact failing line or tool output.
  • If Firefox, Claude Code, OpenCode, Codex, or token-consuming checks are required, state that explicitly before running them.

Test tiers

Tier 0: repository baseline

Run:

bash scripts/check.sh

This covers syntax, unit tests, MCP stdio acceptance, WebSocket bridge acceptance, OpenCode package tests, Codex config validation when Codex is installed, and launch helper tests.

If this fails, diagnose the failing component before moving to higher tiers.

Tier 1: local channel telemetry

Use the active FoxCode MCP server when available:

  • Call status.
  • Expect a live port in 8787..8886, a password, serverVersion, projectDir, and connectedClients.
  • connectedClients == 0 is acceptable before browser launch, but not after a launch skill reports ready.

If the MCP tool is unavailable, use the relevant host diagnostics:

codex mcp get foxcode
opencode mcp list
claude mcp list

Tier 2: extension connection and browser round trip

Launch with the project-profile skill (foxcode-run-project-profile) unless the user specifically asks for the user profile.

After launch:

  • Call status again.
  • Require connectedClients > 0.
  • Run one harmless evalInBrowser smoke:
await api.navigate("https://example.com");
return await api.getTitle();

Expected title: Example Domain.

Then run a DOM smoke:

await api.navigate("https://example.com");
const text = await api.$("h1");
return text.textContent;

Expected result: Example Domain.

Tier 3: popup/session behavior

Use this tier when the change touches the extension background, popup, badge, sessions, reconnection, or eval log.

Verify:

  • Popup receives tool_use and tool_result entries after evalInBrowser.
  • Badge count increments for requests and resets when the popup opens, if badge behavior is in scope.
  • Re-running the launch skill is idempotent: same port stays ready; stale PID on a different port is replaced.
  • Multi-session behavior only when explicitly in scope: two MCP servers, one extension, both sessions visible.

Tier 4: real IDE acceptance

Use this when validating a release candidate or changes to Claude Code, OpenCode, Codex, MCP startup, or launch skills.

Run:

bash scripts/test-ide.sh

This uses real Claude Code, real OpenCode, real Codex, and real Firefox. It may consume model tokens and takes longer than the baseline.

If it fails, inspect the generated acceptance logs first, then classify the failure:

  • IDE startup/config failure.
  • MCP server startup failure.
  • Firefox/extension launch failure.
  • WebSocket connection failure.
  • evalInBrowser behavior failure.
  • External environment failure.

Report format

Return a concise verdict:

  • pass: all requested tiers passed.
  • blocked: a required prerequisite or external environment was unavailable.
  • fail: a FoxCode behavior regressed.

Include:

  • Tiers run.
  • Evidence commands and key output.
  • Root cause if known.
  • Next action, only if there is a concrete one.

Keep looking

Skills are one crate of 608,666. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.