name	smoke-testing-snippets
description	Use when smoke-testing opencode-snippets behavior, especially PTY-driven OpenCode sessions, snippet UX, and end-to-end snippet features.
license	MIT

smoke-testing-snippets

Use this skill before spawning opencode to verify snippet behavior end to end.

Ambient Rules

These are always relevant. Do not hide them behind progressive disclosure.

Rebuild first with bun run build. Local file:///.../opencode-snippets TUI loads built dist output, so stale builds give fake negatives.
For non-interactive sanity checks, prefer opencode run "..." before burning time on full TUI sessions.
For real TUI checks, use pty_spawn plus pty_snapshot or pty_snapshot_wait, not pty_read.
Wait for a stable anchor like autoedit before sending prompt input.
Be patient with keypresses. Use separate pty_write steps with tiny waits when verifying normal UX. One-burst #query + arrow + enter/tab writes are harsher than human typing and should be treated as a separate stress case.
Treat PTY OpenCode sessions as disposable on Windows. Bun segfaults and exit code 3 are host noise unless a calmer repro shows the same bug.
PTY is bad at mouse-hover validation. If a bug only appears in desktop interaction, suspect hover-state or synthetic mouse events.
Smoke test #skill(...) both directly and through snippets that expand into #skill(...).

When debugging PTY-driven OpenCode sessions more deeply, please read PTY + TUI learnings.

For plugin internals, load the opencode-plugin-dev skill via the skill tool.

Fast Path

Spawn OpenCode and wait for autoedit.
Verify normal behavior with paced pty_write steps first.
Then try harsher burst writes to separate real UX bugs from PTY ordering artifacts.
If counts look wrong, inspect matcher output directly before blaming rendering.
For non-TUI snippet features like #skill(...), inspect stored session data too. Visible placeholder text and part metadata can prove transform happened, but they do not alone prove model-facing injected context worked.

What This Skill Helps With

snippet filtering mismatches between expected and actual hits
keyboard navigation vs mouse-hover fights in filtered dropdowns
PTY/TUI verification flow for #, #o, #oc, arrows, tab, enter
separating matcher bugs from render or scroll-state bugs
regular chat and opencode run smoke tests for snippet features like #skill(...)
checking stored session/export data when visible transcript differs from model behavior

smoke-testing-snippets

Use this skill before spawning opencode to verify snippet behavior end to end.

Ambient Rules

These are always relevant. Do not hide them behind progressive disclosure.

Rebuild first with bun run build. Local file:///.../opencode-snippets TUI loads built dist output, so stale builds give fake negatives.

For non-interactive sanity checks, prefer opencode run "..." before burning time on full TUI sessions.

For real TUI checks, use pty_spawn plus pty_snapshot or pty_snapshot_wait, not pty_read.

Wait for a stable anchor like autoedit before sending prompt input.

Be patient with keypresses. Use separate pty_write steps with tiny waits when verifying normal UX. One-burst #query + arrow + enter/tab writes are harsher than human typing and should be treated as a separate stress case.

Treat PTY OpenCode sessions as disposable on Windows. Bun segfaults and exit code 3 are host noise unless a calmer repro shows the same bug.

PTY is bad at mouse-hover validation. If a bug only appears in desktop interaction, suspect hover-state or synthetic mouse events.

Smoke test #skill(...) both directly and through snippets that expand into #skill(...).

When debugging PTY-driven OpenCode sessions more deeply, please read PTY + TUI learnings.

For plugin internals, load the opencode-plugin-dev skill via the skill tool.

Fast Path

Spawn OpenCode and wait for autoedit.

Verify normal behavior with paced pty_write steps first.

Then try harsher burst writes to separate real UX bugs from PTY ordering artifacts.

If counts look wrong, inspect matcher output directly before blaming rendering.

For non-TUI snippet features like #skill(...), inspect stored session data too. Visible placeholder text and part metadata can prove transform happened, but they do not alone prove model-facing injected context worked.

What This Skill Helps With

snippet filtering mismatches between expected and actual hits

keyboard navigation vs mouse-hover fights in filtered dropdowns

PTY/TUI verification flow for #, #o, #oc, arrows, tab, enter

separating matcher bugs from render or scroll-state bugs

regular chat and opencode run smoke tests for snippet features like #skill(...)

checking stored session/export data when visible transcript differs from model behavior

smoke-testing-snippets