gm-verify

name	gm-verify
description	Mechanical verification of the built game: headless build, unit tests, lint, static checks. Explicit invocation only — use /gm-verify.
disable-model-invocation	true

GodotMaker Verify

$ARGUMENTS

You are performing mechanical verification of a built Godot game project. This is a non-creative, checklist-driven process.

Session Setup

FIRST ACTION — before anything else: Write verify to .godotmaker/current_role.

Permission: Read-only with three exceptions — you may write .godotmaker/current_role, append to .godotmaker/stage.jsonl, and write .godotmaker/verify_report.json. Verify never modifies game code or planning docs.

Resume Check

Read .godotmaker/stage.jsonl (treat as empty if missing) — each line is {"role": X, "ts": Y}.

• If no event with role == "build" AND no event with role == "fixgap" exists anywhere in the file → STOP. Tell user to run /gm-build first.
• If the last event has role == "verify" → STOP. Tell the user:

  "Verify already ran at {timestamp} with no state-changing event since. Recommended next: /gm-evaluate. If you need to redo this step or have other plans, just tell me."

• Otherwise → proceed (verify is naturally re-invoked after each build/fixgap cycle).

Run the checks

From the project root:

python tools/run_verify.py

run_verify.py wraps the four mechanical checks (build / unit tests / lint / static check) and prints a JSON document matching Output Format Section B to stdout. Capture stdout.

What the script does:

• Reads godot_path from .claude/godotmaker.yaml; falls back to plain godot from PATH. A missing or broken binary surfaces as a tooling_notes[].suggested_fallback = "escalate" entry.
• Runs <godot_path> --headless --quit and writes blocking Godot diagnostics to checks.build.errors[].
• Runs <godot_path> --headless ... res://addons/gdUnit4/bin/GdUnitCmdTool.gd --ignoreHeadlessMode --add res://test/ --report-directory <temp> and parses the generated JUnit XML into checks.unit_tests.{passed, failed, failures[]}. Stdout is diagnostic fallback only.
• Stubs checks.lint as pass with format_drift: null. Do NOT re-enable here.
• Delegates checks.static_check to python tools/check_project.py <project_dir> --build --ecs --tests --plan --mcp.
• Exit code 0 = ran to completion (per-check pass/fail is in the JSON).

Sanity-check the script output

Before writing the report, validate. Block on any of these:

• JSON does not parse, or any of result / ts / checks / tooling_notes is missing
• Any of the four checks.{build,unit_tests,lint,static_check} entries is absent
• result == "pass" but tooling_notes is non-empty — re-run or escalate
• checks.unit_tests.passed + .failed == 0 — spot-check by running the gdUnit4 command directly with --report-directory <temp>
• Any tooling_notes entry whose crashed_on looks unrelated to the failing check

If any block fires, diagnose by running the implicated command yourself, then either re-run run_verify.py or surface the issue verbatim to the user. Do NOT silently rewrite the script's output.

Output Format

You produce two outputs:

A. Human-readable report (chat)

Build this from the JSON the script returned — do not re-run any command for the chat side.

## Verification Report

### Build
Result: PASS | FAIL
{If FAIL, one line per checks.build.errors[] entry: `- {file}:{line}: {message}` (file/line may be empty)}

### Unit Tests
Result: PASS | FAIL
{N passed, M failed}
{If FAIL, one line per checks.unit_tests.failures[]: `- {test}: {message}`}

### Lint
Status: SKIP (gdtoolkit disabled — ROADMAP R-112)

### Static Check
Result: PASS | FAIL
{If FAIL, one line per checks.static_check.issues[]: `- {check}: {detail}`}

### Overall: PASS | FAIL

{If tooling_notes is non-empty, append:
## Tooling Notes
- {tool}: {error} (suggested_fallback: {suggested_fallback})
…}

B. Machine-readable report (.godotmaker/verify_report.json)

Write this file every run (PASS or FAIL). /gm-build and /gm-fixgap read it on their next invocation to translate failures into pending tasks.

Schema:

{
  "result": "pass | fail",
  "ts": "<UTC ISO 8601 timestamp, e.g. 2026-05-07T14:23:00Z>",
  "checks": {
    "build": {
      "result": "pass | fail | error",
      "errors": [
        {"file": "src/foo.gd", "line": 42, "message": "Identifier 'bar' not declared"}
      ]
    },
    "unit_tests": {
      "result": "pass | warn | fail | error",
      "passed": 624,
      "failed": 0,
      "failures": [
        {"test": "test_player_input::test_jump", "message": "expected 10, got 0"}
      ],
      "warnings": [
        "Found 4 possible orphan nodes."
      ]
    },
    "lint": {
      "result": "pass | warn | fail | error",
      "issues": [
        {"file": "src/foo.gd", "rule": "max-line-length", "message": "line too long"}
      ],
      "format_drift": {
        "file_count": 92,
        "command": "gdformat src/ test/ scenes/"
      }
    },
    "static_check": {
      "result": "pass | fail | error",
      "issues": [
        {"check": "missing_unit_test", "detail": "s_level_up_overlay has no test"}
      ]
    }
  },
  "tooling_notes": [
    {
      "tool": "gdlint",
      "crashed_on": "src/foo.gd",
      "error": "NotImplementedError at gdtoolkit/linter/class_checks.py:144",
      "suggested_fallback": "exclude_file",

      "narrowed_command": null,
      "rule_name": null,
      "check_name": null
    }
  ]
}

Field rules:

• Top-level result — "pass" iff every checks.*.result ∈ {pass, warn}. Any fail / error makes overall fail. tooling_notes alone never makes overall fail — the error it pairs with does.

• ts — UTC ISO 8601 at the moment you write the file. Consumers compare it against their own last-event timestamp for freshness.

• All array fields are required (possibly empty []). Do not omit them.

• Per-check result — pass / fail are project-content. warn is non-blocking diagnostic noise (lint style drift or gdUnit warnings such as orphan nodes when every assertion passed). error means the tool itself crashed and the project's actual state is unknown for this check; pair error with exactly one tooling_notes entry. Consumers fix error via config, NOT project code.

• format_drift — object when gdformat --check reports drift; null otherwise.

• suggested_fallback + matching operand — the producer fills the operand so the consumer can act deterministically:

  suggested_fallback	Required operand
  exclude_file	crashed_on (already required on every note)
  scope_narrow	narrowed_command (replacement command, e.g. "gdlint src/")
  add_gdlintrc_rule	rule_name (e.g. "class-name")
  skip_check	check_name (e.g. "missing_unit_test")
  escalate	— (none)

  Producer rule: if you cannot fill the required operand for a non-escalate fallback, emit escalate instead.

  Consumer rule (open-enum forward-compat): a missing required operand or an unknown suggested_fallback value MUST be treated as escalate (surface to user, do NOT auto-fix). Never crash.

On Failure

When the script's JSON has result: "fail":

1. Write the script's JSON verbatim to .godotmaker/verify_report.json.
2. Emit the chat report (Section A) and tell the user which checks failed. Suggest /gm-build if the last state-changing event was build, /gm-fixgap if it was fixgap.
3. Do NOT append a verify event to stage.jsonl — only PASS records a stage event.

When Done

When the script's JSON has result: "pass":

1. Write the script's JSON verbatim to .godotmaker/verify_report.json. (Field rules apply: tooling_notes == [], all checks.*.result ∈ {pass, warn} — the script enforces these but spot-check them once more before writing.)
2. From the project root run python tools/append_stage_event.py verify to append a {"role": "verify", "ts": "<server-generated UTC>"} line to .godotmaker/stage.jsonl. Do NOT hand-write the JSON or the timestamp — the helper exists so the timestamp comes from the system clock, not your own output.
3. git add -A && git commit -m "chore(verify): <Tag>"
4. Inform the user: Verify complete. Recommended next: /gm-evaluate