Running Tests

Running Tests for opencode-swarm

This skill is about executing tests safely. For writing tests, see writing-tests.

Graph-first evidence contract

Use repo_map test_pack only to discover focused candidate tests; Bun/shell output remains execution authority. Graph evidence is advisory only. If freshness is stale or inconclusive, confidence is low, source is missing, the language is unsupported/dynamic, the graph is absent, or the action fails, select tests from direct source, imports, and repository conventions.

⛔ The Scope Rule That Prevents Session Kills

convention discovery accepts one source file at a time (or explicit direct test files). graph and impact discovery accept a bounded normalized source array of up to MAX_SAFE_TEST_FILES = 50. Do not exceed that input cap or respond to scope_exceeded by widening to scope: 'all'.

The final unique resolved-test cap still binds for every discovery scope: when resolution produces more than 50 test files, test_runner returns scope_exceeded without executing them. Split a larger graph/impact selection into intentional bounded batches.

Three-Layer Defense Against Session Blocking

test_runner bounds source selection and resolved test execution before a session can fan out without limit:

Layer 1 — Scope-specific normalized-input guard

convention rejects more than one source file for convention discovery. graph and impact reject more than MAX_SAFE_TEST_FILES = 50 normalized source files before fan-out. Explicit direct test files remain allowed for convention scope.

Layer 2 — Advisory resolution estimate

For graph and impact, estimateFanOut(sourceFiles, workingDir) reads the cached impact map and reports a bounded, advisory count of unique candidate tests without spawning subprocesses. Resolution metadata preserves whether the estimate was advisory or unavailable and any cache status; it is not a substitute for the final cap.

Layer 3 — Bounded traversal + final unique-test check

Graph and impact traversal are bounded to the safe budget and report scope_exceeded when the budget is exceeded. After fallback, normalization, and deduplication, the final unique testFiles.length is compared with MAX_SAFE_TEST_FILES; an excess returns scope_exceeded before execution.

Result: When fan-out exceeds the safe threshold, the session gets outcome: 'scope_exceeded' instead of hanging.

Decision Tree: test_runner tool vs bun shell command

Do you need to run tests?
│
├─ Single test file, targeted validation
│   └─ Either works. Prefer shell: bun --smol test <file> --timeout 30000
│
├─ Multiple test files in the same directory (e.g. all agents tests)
│   └─ Shell only — per-file loop. These are explicit test files, not graph/impact sources.
│
├─ Find tests related to ONE OR MORE changed source files (up to 50 normalized files)
│   └─ test_runner is fine: { scope: 'graph', files: ['src/agents/coder.ts', 'src/tools/test-runner.ts'] }
│      (graph/impact input is bounded; final unique resolved-test cap still applies)
│
├─ Find tests related to MORE THAN 50 changed source files
│   └─ Split into intentional bounded graph/impact batches or use a shell loop.
│      Do not use scope:'all' as a fallback.
│
└─ Validate the entire repo (pre-push)
    └─ Shell only — 5-tier suite from commit-pr skill. Never test_runner scope:'all'.

Scope Safety Reference

Rule of thumb: Pass one source file to convention; pass a normalized array of at most 50 source files to graph or impact. Use a shell loop or intentional batches when the source selection or final resolved test set exceeds 50.

For one named Go or CTest test, bypass file discovery with an exact native selector:

• Go: { scope: "target", native_target: { framework: "go-test", name: "TestName[/Subtest]", path: "relative/package" } }
• CTest: { scope: "target", native_target: { framework: "ctest", name: "ExactTestName", path: "relative/build-dir" } }

The target name is treated literally, the directory must stay within the project root, and the runner never falls back to a broader package or build-tree sweep.

Per-File Isolation Loops

CI runs agents/tools/services in per-file isolation (one bun --smol process per file). Reproduce this locally with the following loops.

bash (Linux / macOS)

# Single directory — per-file isolation
for f in tests/unit/agents/*.test.ts; do
  bun --smol test "$f" --timeout 30000
done

# Multiple directories
for dir in tests/unit/tools tests/unit/services tests/unit/agents; do
  for f in "$dir"/*.test.ts; do
    bun --smol test "$f" --timeout 30000
  done
done

# Stop on first failure (useful for debugging)
for f in tests/unit/agents/*.test.ts; do
  bun --smol test "$f" --timeout 30000 || { echo "FAILED: $f"; break; }
done

PowerShell (Windows)

# Single directory — per-file isolation
Get-ChildItem tests/unit/agents/*.test.ts | ForEach-Object {
  bun --smol test $_.FullName --timeout 30000
}

# Multiple directories
@('tests/unit/tools', 'tests/unit/services', 'tests/unit/agents') | ForEach-Object {
  Get-ChildItem "$_/*.test.ts" | ForEach-Object {
    bun --smol test $_.FullName --timeout 30000
  }
}

# Capture output (avoids truncation on large output)
Get-ChildItem tests/unit/agents/*.test.ts | ForEach-Object {
  bun --smol test $_.FullName --timeout 30000
} | Out-File "$env:TEMP\test_out.txt"
Get-Content "$env:TEMP\test_out.txt" | Select-Object -Last 50

Common PowerShell pitfalls:

• for f in ...; do — invalid, use Get-ChildItem | ForEach-Object
• Select-String -Last N — invalid parameter, use Select-Object -Last N
• 2>&1 2>&1 — duplicate redirection, causes parse error; use 2>&1 once
• && — not supported in PowerShell 5.1; use ; if ($?) { cmd2 } instead
• bun test --exec bash — fails on Windows hosts with ENOENT (bash is not available in standard PowerShell). Use bun test directly or a PowerShell-based loop instead.
• After bun install --frozen-lockfile --force, non-elevated Windows shells can hit EPERM while reading refreshed node_modules entries. Treat that as a host permission/access issue: rerun the same focused Bun command with approved/elevated access before diagnosing it as a code or test failure.

Batch vs Per-File: Which Directories Need Isolation?

Truncated Output Recovery

When bun test output exceeds the bash tool's buffer, it is saved to a file with an ID like tool_dff778.... This ID format is not accepted by retrieve_summary (which only reads S1, S2 etc. format IDs). The output is effectively lost.

Prevention — pipe to a file explicitly:

# PowerShell
bun --smol test tests/unit/agents --timeout 60000 |
  Out-File "$env:TEMP\test_out.txt"
Get-Content "$env:TEMP\test_out.txt" | Select-Object -Last 50

# bash
bun --smol test tests/unit/agents --timeout 60000 2>&1 | tee /tmp/test_out.txt
tail -50 /tmp/test_out.txt

To get a clean pass/fail summary only, filter immediately:

# PowerShell — show only summary lines
bun --smol test tests/unit/agents --timeout 60000 |
  Select-String "pass|fail|error" |
  Select-Object -Last 10

# bash
bun --smol test tests/unit/agents --timeout 60000 2>&1 | grep -E "pass|fail|error" | tail -10

Verifying Pre-Existing Failures

Before documenting a failure as "pre-existing," prove it exists on main without affecting your working tree. Use a Git worktree — safer than git stash (stash can drop untracked files, fail on locked files on Windows, and leave you in an inconsistent state).

# bash — create a throwaway checkout of main
git worktree add /tmp/repro-check origin/main
bun --smol test /tmp/repro-check/tests/unit/agents/architect-workflow-security.test.ts --timeout 30000
git worktree remove /tmp/repro-check

# PowerShell — same pattern (use Join-Path for robust separator handling)
git worktree add "$env:TEMP\repro-check" origin/main
$testPath = Join-Path "$env:TEMP\repro-check" "tests\unit\agents\architect-workflow-security.test.ts"
bun --smol test $testPath --timeout 30000
git worktree remove "$env:TEMP\repro-check"

Decision after checking:

• Fails on main too → pre-existing. Document under ## Pre-existing failures in PR body. Continue.
• Fails only on your branch → you introduced it. Fix before pushing.

⚠️ Check your own session history first. Before documenting anything as pre-existing, confirm you did not fix or update this test earlier in the current session. A test you fixed 20 messages ago is not pre-existing — listing it as such in the table or PR body is incorrect and will be caught in review.