code-test

code-test

This is a Codex-native Skill projection generated from the portable capability contract. It is adapter-owned output, not a legacy compatibility Skill copy.

Source

• Portable source: capabilities/code-test.md
• Runtime check: adapters/codex/bin/preflight.sh capability-info code-test
• Bootstrap: adapters/codex/AGENTS.md

Use

1. Read capabilities/code-test.md for the runtime-neutral contract.
2. Run adapters/codex/bin/preflight.sh capability-info code-test.
3. Obey the reported status:
   • instruction-only: use this Skill as Codex guidance plus explicit preflight guards.
   • tool-contract: report the named tool_contract, run any tool_contract_check, and obey runtime_surface / fallback before claiming full support.
   • unsupported: stop or use the reported fallback.

Shape

• Identifier: code-test
• Invocation class: parent-invoked
• Supported modes: none
• Argument shape: <plan name, path, or test scope> [--intensity direct|quick|standard|strong|thorough|adversarial]
• Portable meaning: Verify implementation results in stages and record evidence.

Portable Contract

• Invocation semantics: Assess graduated verification after code-execute or on demand to verify code correctness. Intensity-derived rigor scales final verification and test-adequacy review; it does not force a separate parallel QA loop by itself. The capability resolves a plan path, changed-file list, or test scope, assesses the applicable test levels, runs only checks whose evidence is missing or invalidated by the change, stops on the first failing new check, and records durable evidence before reporting a verdict. When the verification target includes a report spectrogram, assess the fail-closed figure semantic verifier against its manifest and report. Reuse its result only while the manifest, report, PNG hash, and visual-review evidence still match. Missing exact 48 kHz full-band metadata, range-compatible claims, shared-scale evidence, or a hash-current visual review is a failed level. For a planned QA leg with sealed done_when items, its <primary-report>.items.json is a derived output of that exact primary report. The existing report scope, sub-session fence, state ledger, and continuation carry the sidecar with the primary; no separate path input is needed. Adapters may expose this capability through native commands, skill files, prompt instructions, or explicit wrappers. The adapter must report unsupported runtime mechanics instead of silently treating another runtime's native file format as portable. Put pytest basetemp and tool scratch data under the supplied runtime_dir or system temporary storage. Keep test reports and compact evidence in the declared artifact outputs.

Projected Portable Details

Artifact Ownership

Artifact root: core/CONVENTIONS.md §5.1.

When invoked from a standard+ autopilot-code stage cycle, write verification evidence only to the test node's declared outputs and write scope. Return the final verdict to code-report, which owns pipeline_summary.md; the code-test stage must not write that report-owned artifact. Standalone invocations should create or reuse an appropriate plans/<date>_<slug>/ work-cycle directory and may update their own standalone summary when no code-report stage exists.

Required evidence:

• test target resolution: plan path, changed-file list, or inferred scope;
• command log or explicit skip/block reason for each attempted level;
• failing level and first actionable error when verification fails;
• final one-line verdict suitable for handoff to code-report.

Artifact Producer Lifecycle

code-test is a standard+ stage worker: it never issues its own campaign or cycle. It receives the owner's open cycle through AGENT_ARTIFACT_CAMPAIGN_ID/AGENT_ARTIFACT_CYCLE_ID/AGENT_ARTIFACT_PRODUCER_ID/ AGENT_ARTIFACT_CYCLE_DIR/AGENT_ARTIFACT_OUTPUT_DIR (dispatch env pass-through); the resolver applies the test node's declared outputs and write scope. artifact_producer.py owns the open cycle and its output paths; finalize and admit-shared belong to the owner, never to a stage worker. See producer_lifecycle in capabilities/topologies.json.

Role Requirements

Use portable role names from roles/README.md and core/CONVENTIONS.md. Concrete model names, subagent frontmatter, and runtime-specific tool lists belong in adapter files.

Minimum role mapping:

• verification: QA role using roles/units/qa/test.md;
• review: optional QA reviewer for test adequacy when selected by QA/intensity risk;
• reporting: editorial/reporting role only for user-facing summary polish, not for changing the test verdict.

Guard Requirements

Adapters must preserve the portable invariants relevant to this capability:

• resolve artifact root through utilities/artifact-root.sh or equivalent logic;
• use DB memory paths, not runtime-native memory files.

Additional test-entry gates:

• run roles/units/qa/test.md semantics or the adapter-native projection of that mode before claiming verification;
• if the adapter reports a verification-runner tool contract, run its contract check or report unavailable;
• do not modify source files while acting in code-test;
• do not claim independent QA review unless a separate QA role, headless worker, or external reviewer actually ran.

Portable Procedure

1. Resolve the verification target:
   • if a plan path is provided, read the plan's verification section and the corresponding checklist/changed-file evidence;
   • if changed files are provided, use them directly;
   • otherwise infer recent changed files from git state and report the inference.
2. Assess the applicable levels from roles/units/qa/test.md: syntax, import, smoke, functional, integration, and behavioral runtime observation for user-facing surfaces. These levels classify evidence; they do not require a restart at syntax when valid evidence already covers a level. Use existing execution, review, test, and CI artifacts when their source, scope, inputs, dependencies, and run conditions still apply. Identify the changed delta, any known failure, and any required unverified dimension. If changed outputs include a spectrogram report, assess the fail-closed figure semantic verifier against its manifest and report, reusing its result only when the evidence remains current.
3. Run only selected checks whose evidence is missing or invalidated, in level order, and stop at the first failure. Preserve known unresolved failures; after a fix, rerun the check affected by that fix. A handoff or a historical PASS marker alone neither invalidates command evidence nor satisfies the current test gate.
4. Record commands, outputs or excerpts, skips, blockers, and the first actionable failure in test_logs/.
5. Emit the final verdict as a handoff to code-report; in a standard+ stage cycle, code-report alone updates pipeline_summary.md. When no code-report stage exists, a standalone invocation may update its own standalone work-cycle summary.
6. Return a concise report path plus verdict to the caller.

Shared Procedure References

Read the complete portable procedure files listed here when they apply; this logical read-set is shared across Claude, Codex, and OpenCode.

• capabilities/code-test/references/procedure.md

Workflow Evidence

• Before capability grounding/spec-changing work: adapters/codex/bin/preflight.sh route code-test [cwd] [session-id]
• For workflow state: adapters/codex/bin/preflight.sh status [cwd] [session-id] and adapters/codex/bin/preflight.sh prompt-signal [cwd] [session-id]

Do not use legacy compatibility Skill files or non-native adapter Skill files as Codex-native source. Those files are compatibility/reference surfaces only.