test

Run the fast automated gate for the current change.

When to invoke

Use for $test, “test my changes”, or a quick pre-save validation. Use qa for broader behavioral proof and adversarial fixtures.

Arguments: $ARGUMENTS — optional paths, --all, or explicit --graph.

Default gate

1. Run bash bin/base-branch.sh to print the configured integration branch. Scope to supplied paths, --all, or changed files against that branch.
2. Run static analysis once:

bash bin/test-changes.sh <scope>

--all scans source skills plus Runtime-neutral shell adapters. Do not treat a missing historical .claude/commands directory as test coverage.

3. Run the architectural ratchet once (source instances only — installed instances do not ship the distribution engine and skip this step):

if [ -f bin/capability-distribution.mjs ] && [ -f capability-distribution.json ]; then
  bash bin/node-run.sh bin/capability-distribution.mjs runtime-skill-audit \
    --root . --strict-contract
fi

This checks the complete skill denominator, QMD/graph/control-plane/write coupling ceilings, and migrated-skill contracts.

4. Select the smallest relevant existing test suites from the changed paths and skill-references.json. Run independent suites in parallel where the harness supports it. Do not rerun the same suite through multiple rituals.
5. For changes to cross-harness Runtime sources or skills, run Codex/Pi/Prime bundle parity checks.

Optional graph validation

Only when the user explicitly passed --graph or changed the graph projection adapter itself:

• require Connected graph capability;
• extract bounded read-only Cypher fixtures from the named graph files only;
• reject CREATE, MERGE, SET, DELETE, REMOVE, or DROP;
• validate through the optional graph projection adapter;
• report graph unavailable as an optional-check gap, not a Local test failure.

Never scan every skill for Cypher and execute it against a live database.

Result

Render one compact pass/fail report with:

• static checks;
• Runtime architecture audit;
• targeted suites and counts;
• bundle parity when applicable;
• optional graph check only when requested;
• exact failures and the safe next action.

Warnings do not block unless they represent a privacy, authorization, canonical-order, or destructive-action risk.

Rules

• No default Neo4j, API, or network dependency.
• Tests must not modify production/live QMD, graph, control-plane, or canonical memory state.
• Use isolated fixtures for writes and destructive/failure behavior.
• Never call graph functionality the correct default Runtime boundary.
• Do not claim “safe to save” when required suites were skipped.

Emit content-free telemetry after the gate:

bash bin/telemetry.sh emit "command" '{"command":"test"}' 2>/dev/null &