godot-runtime-verify

name	godot-runtime-verify
description	Verify a built Godot 2D scene at runtime: run the scene (press-play or headless), read the captured debugger output, classify any runtime errors against a Godot-specific taxonomy, and decide a PASS/FAIL runtime gate for the slice's acceptance behavior. The run's real output IS the Layer-1 runtime evidence (ADR-0048); a claimed-but-not-shown run is unverified. MCP-agnostic about how the scene is run; it verifies and routes fixes, it does not apply them. Use after a Godot slice is implemented to gate runtime behavior the static checks (lint, typecheck) cannot catch. Do not use to plan a scene (use godot-scene-plan), to write or fix the code (use implement-approved-slice or implement-slice-complement), to triage a failure into a fix size (use incident-triage), or with no implemented scene to run.
metadata	{"category":"execution-and-closure","primary-cursor-mode":"Agent","multi-repo-aware":false,"context-layers-consumed":["memory"],"context-layers-produced":["memory"],"tools":["Read","Write","Edit","Bash","Glob","Grep"],"x-wos-profiles":["full"],"provenance":"first-party","suggested-model":"claude-sonnet-4-6"}

Act as a senior Godot engineer running a built 2D scene and verifying its runtime behavior before the slice is closed.

Goal: Run the implemented Godot scene (press-play in the editor or a headless run), capture and read the debugger output, classify any runtime errors against a Godot-specific taxonomy, and decide a PASS or FAIL runtime gate for the slice's acceptance behavior. This is the "feedback edge" that the static checks cannot cover: most Godot bugs are runtime bugs a linter never catches (EXTERNAL_RESEARCH.md A2). The command's verdict is a Layer-1 runtime gate per the three-layer model (wos/gate-conditions.md, ADR-0048): the run's actual output is the evidence, and it feeds Layer 2 (review-hard) and Layer 3 (human approval), never replacing them. The command verifies and routes; it does not write or fix code.

Mandatory context bootstrap (before any output):

• Read these sections in WORKFLOW_OPERATING_SYSTEM.md first:
  • ## LLM execution contract
  • ## Editor mode policy (mode definitions only; the tool mapping table is lazy-loaded in wos/editor-mode-mappings.md and needed only for non-Claude-Code tools)
  • ## Global output contract (including Adaptive handoff and Mode selection rule)
  • ## Cross-cutting workflow guardrails
• Bootstrap tiers (ADR-0025): the light-weight commands (branch-commit, what-next, where-we-at, slice-closure, compact-task-memory) may skip ## Editor mode policy good-fits lists and ## Cross-cutting workflow guardrails sequencing heuristics, reading only the mode definitions and the core guardrail rules (routing memory, command-less input triage, official command names, material change, no-op). The full tier is measured at 9610 tokens: the combined size of the four always-read WORKFLOW_OPERATING_SYSTEM.md sections listed above. The reduced tier is a self-declared estimate of about 3,500 tokens for the trimmed subset above; it has not been independently re-measured by the same method, and should be read as an estimate rather than a fresh figure. The same reduced tier extends to the high-frequency execution commands implement-approved-slice and sync-task-state (v3 wave1 item D: the most-invoked commands pay the bootstrap most often; state-reconcile deliberately stays on the full tier, cross-artifact judgment needs the full guardrail context).
• Cache-amortized layer (ADR-0006): this bootstrap floor is a cache-amortized cost, not a per-command tax paid in full on every invocation. It sits in the prompt cache for the session and is paid at write cost once per cache TTL window, then at roughly 0.1x on cached reads inside that window. Account for it separately from any per-skill Load budget (the generated .claude/skills/<name>/SKILL.md body); the two are different layers and should not be summed into one figure.
• Session bootstrap reuse (skip-if-unchanged; v3 wave1 item D): WHEN this same conversation already performed this bootstrap read in an earlier turn that is still VISIBLE in the current context window AND WORKFLOW_OPERATING_SYSTEM.md has not changed since, the command MAY skip the re-read and cite the earlier one instead, emitting one Command transcript line: Bootstrap: reusing turn <N> read, WOS unchanged. This is a scoped exception to the context-budget re-fetch rule (wos/context-budget.md, "The re-fetch rule"), justified because the bootstrap sections are one large, static, byte-identical read repeated every turn rather than a variable tool result; the re-fetch rule still governs every other tool result without exception. VISIBLE means the bootstrap section text itself is still present and quotable in the window right now, not merely that the record of an earlier read exists. On a harness that clears, a tool result can be emptied while the record that the tool ran survives (ADR-0114); a command that finds only that record, without the section text still readable, has not satisfied VISIBLE and must re-read. Self-declared memory after a compaction never qualifies (re-read instead), and a stateless-per-turn harness is excluded. The auditable-skip rule applies: the transcript line is mandatory; a silent skip is invalid output.
• Read additional sections only when relevant to this command's role.
• Read the commands/ directory command inventory to ensure command names and availability are current.
• Align all routing recommendations and next-command suggestions with the current command set.
• Official next-command names only: every recommended next command (including the handoff Run now line) MUST be the basename of an existing commands/<name>.md file in this workflow repository. Never invent names.

Required inputs:

• active task folder path
• the implemented slice or feature under verification, and its acceptance behavior (the observable outcome that means it works, ideally the slice's EARS exit criterion)
• how the scene was run and the real captured output: the run mechanism (an MCP server run tool, the Godot CLI headless run, or a human pressing play) plus the actual debugger or console output from that run. When the output is not yet captured, this command STOPS and asks for it rather than asserting a result.
• the target Godot version, when relevant to interpreting an error

Operating rules:

• Do not write or fix code; this command runs and verifies, then routes a fix to the right command. Within-scope tidying of the report is allowed.
• K.2 scope note (P2-8, dogfood-wave-2 2026-07-12): this command's own artifact (GODOT_RUNTIME_VERIFY.md) is outside the K.2 11-file substrate scope (commands/_shared/substrate-write-protocol.md) and needs no transaction header. Update TASK_STATE.md's ## Last completed step afterward as ordinary operator hygiene.
• Handoff: end with the adaptive ### Handoff block per WORKFLOW_OPERATING_SYSTEM.md ## Global output contract (Mode A compact or Mode B full).
• Evidence, not trust (ADR-0048): the run's actual output (debugger or console log) MUST be shown. A run whose result is claimed but whose output is not shown is unverified, exactly like an asserted "tests pass". Do not fabricate runtime output; if it was not captured, STOP and ask for it.
• MCP-agnostic (DECISIONS D-1): the command names no specific MCP server. Whatever ran the scene (an MCP run tool, the Godot CLI --headless, or a human) is the operator's choice; the command verifies the output, it does not prescribe the runner.
• Verify, then route the fix; do not fix here. A FAIL routes to incident-triage (to size the fix when the cause is unclear) or implement-slice-complement (for a bounded known fix inside the slice intent). Reopening a signed-off decision routes to post-review-pivot.
• Bounded retry (wos/gate-conditions.md interactive bounded retry). When this gate is used in a hold-until-pass loop (re-run after each fix), cap consecutive failed runs at a small N (default 3 to 8). On reaching the cap, STOP and escalate to the human rather than looping; a hold-until-pass note without the cap reintroduces the infinite-retry loop.
• Layer placement: a PASS here is Layer-1 runtime evidence; it does not skip Layer 2 (review-hard, repo-consistency-sweep) or Layer 3 (human approval).
• No-op rule: if a current runtime verification already covers this slice with no material change (the scene and acceptance behavior are unchanged since the last PASS), return a short NO_OP note and route forward.
• Environment preflight. Before the gate runs, the command MUST resolve the Godot binary: try godot --version on PATH first; on macOS, fall back to the app bundle binary (/Applications/Godot.app/Contents/MacOS/Godot). Record the resolved path in the run evidence. If no binary resolves, STOP and ask the user where Godot lives rather than improvising a runner.
• Persistent probe harness. Probe scenes and scripts live under probes/ in the game repo and are kept under version control, not written and deleted per slice. A probe MUST be self-terminating: call get_tree().quit() on PASS or FAIL, with a physics-frame backstop so a hung probe still exits. Drive behavior by calling handlers directly (for example spawner._drop()), never through simulated input timing or wall-clock waits. Write a probe as a real scene (extends Node, a companion .tscn, invoked headless against that scene), not as a bare --script-invoked SceneTree/MainLoop override: the latter does not have project autoloads registered (a compile error, Identifier not found: <Autoload>, with the process still exiting 0) even though the identical autoload resolves correctly for a real scene-tree node script. After an instant, non-simulated reposition (global_position = ... on a physics body), wait a few physics frames (empirically 3 or more) before asserting an Area2D overlap query, since the physics server's overlap state updates on a deferred schedule, not synchronously with the assignment.
• Adversarial probe requirement. The runtime gate for a mechanic acceptance MUST include at least one adversarial or stress probe (rapid repeated input, boundary states, spam of the core action) alongside the happy-path probe. A gate that ran only happy-path probes is incomplete evidence and MUST say so in its verdict.
• Step 1: Confirm the run mechanism and the acceptance behavior. Restate the slice under verification, its acceptance behavior (the EARS exit criterion when present), and how the scene was run. If the real run output is not provided, STOP and request it (do not proceed on an asserted result).
• Step 2: Read the captured output. Read the debugger or console log from the run. Quote the load-bearing lines (errors, warnings, the absence of expected output) verbatim in the report; do not paraphrase an error.
• Step 3: Classify each runtime observation. Tag every error or anomaly with one taxonomy code: SCRIPT_ERROR (a GDScript runtime error: nil method call, type mismatch, bad cast), MISSING_NODE_OR_RESOURCE (a node path not found or a resource that failed to load), SIGNAL_NOT_CONNECTED (an expected signal never fires or was never wired), NULL_REFERENCE (access to a freed or never-assigned node), PHYSICS_OR_COLLISION (a body that does not move, a collision layer/mask mismatch), INPUT_NOT_MAPPED (an action missing from the input map, or no touch binding on a mobile target), PERFORMANCE_STALL (frame drops or a hang; defer numeric budgets to performance-budget), STATE_INVARIANT_VIOLATION (an observable game-state rule broken with no thrown error and no other code fitting: a double-spend, a double-placement, an impossible state transition; exactly the class the adversarial-probe requirement exists to surface), or CLEAN (no runtime error and the acceptance behavior was observed). One line per observation: the quoted symptom, the code, and the most likely cause.
• Step 4: Verdict per acceptance criterion. For each acceptance behavior, state observed, not-observed, or unverified (output not shown), grounded in the captured log.
• Step 5: Gate decision. PASS only when the scene ran, there is no unhandled runtime error, and every acceptance behavior is observed. Otherwise FAIL (one or more errors or a not-observed behavior) or BLOCKED (output unverified, or the bounded-retry cap was reached). State the decision in one line with its reason.
• Step 6: Write the report. Save as GODOT_RUNTIME_VERIFY.md (or GODOT_RUNTIME_VERIFY_<slice>.md when several slices are verified) in the active task folder: the run mechanism, the quoted output, the classification table, the per-criterion verdict, and the gate decision.
• Step 7: Emit or update the playtest runbook (ADR-0084). Alongside the machine-run gate, write or update PLAYTEST_RUNBOOK.md in the active task folder: how a human launches the scene (the run command or the press-play steps and the main scene to set), the specific things to exercise (the acceptance behaviors plus what the automated gate cannot judge: feel, difficulty, pacing, and fidelity to the reference or the MECHANICS_SPEC.md), and where the playtester's notes go. The runbook is the durable, repeatable counterpart to the automated gate. This gate catches crashes and missing behaviors; the human playtest catches wrong-but-running mechanics the gate passes, which is the ADR-0084 failure: the runtime gate PASSED a core mechanic that was objectively wrong, because it verifies the contract and cannot question it. An improvised one-off run instruction is not a runbook; the artifact is the point.
• Route human playtest notes to pr-feedback-ingest --playtest (ADR-0084). When the operator returns playtest feedback (the game runs but plays wrong, a mechanic is off, a screen flow is missing), the handoff routes it to pr-feedback-ingest --playtest, a first-class corrective ingestion path, not to a general review command. Do not absorb playtest feedback into an ad-hoc review.
• Per-slice adoption (ADR-0084). A Godot slice with runtime-observable behavior runs this gate, or records an explicit skip reason in the slice notes (for example a pure-data or config slice with nothing to run). A silently skipped runtime gate is the observed decay mode (the dogfood ran the gate once and dropped it for the two larger builds); the explicit skip line keeps the decision visible.

Required output:

1. Slice under verification and its acceptance behavior
2. Run mechanism + the quoted real output (or a STOP requesting it)
3. Classification table (symptom, taxonomy code, likely cause)
4. Verdict per acceptance criterion (observed | not-observed | unverified)
5. Gate decision (PASS | FAIL | BLOCKED) with reason
6. PLAYTEST_RUNBOOK.md written or updated (how to run, what to exercise including mechanic feel, where notes go)
7. Recommended next command (the fix route on FAIL, closure on PASS, or pr-feedback-ingest --playtest when the operator returns human playtest notes)

Claim grounding (active epistemic humility)

Claim grounding (active epistemic humility). This block governs what you may assert and how you record it. It is keyed to the substrate section you are writing, not to which command is running, and it is INERT on any output that writes none of the claim-bearing sections below. Full contract and rationale: wos/active-epistemic-humility.md.

1. When this applies. This block fires ONLY while you are writing a claim-bearing substrate section: TASK_STATE.md ## Current known facts, ## Risks to watch, ## Observations, ## Active files in scope, ## Canonical decisions; DECISIONS.md ## Locked decisions; IMPLEMENTATION_PLAN.md ## Current gaps, ## Risks and mitigations; IMPACT_ANALYSIS.md; EXTERNAL_RESEARCH.md; REFERENCES.md; or any section whose content is a statement a later command or a human decision will act on. WHEN your output writes none of these, this block imposes nothing: skip it and proceed. This is the D-13 inert clause; a fully-grounded or claim-free output pays nothing.

2. The unit is the load-bearing claim. A load-bearing claim is one a downstream command or a human decision consumes. A passing aside is not load-bearing; a statement someone will act on is. Apply the rest of this block per load-bearing claim, not per sentence.

3. Ground it or abstain. Before you assert a load-bearing claim, trace it to the enumerable grounded set: a captured REFERENCES.md entry, a file read in this session, command output actually seen, or a passing deterministic gate. A claim supported only by model memory is OUTSIDE the grounded set, including when you are right, because that support is not observable. WHEN a load-bearing claim falls outside the set, do NOT assert it: either investigate until it is grounded, or abstain per rule 6.

4. Status records provenance, never confidence. WHERE you attach an epistemic status to a claim, the status names WHERE THE CLAIM CAME FROM: a REFERENCES.md entry title, a file path plus line, or the gate output it came from. It SHALL NOT express a degree of certainty. Do NOT add a confidence field, a numeric threshold, or a self-assessment prompt anywhere; a self-reported confidence signal is not a usable control signal (wos/active-epistemic-humility.md Part 1.3). A status whose referent slot is empty is read as UNKNOWN, not as a weak yes.

5. Persisted claims carry the status; chat-only claims carry it when they route. Every load-bearing claim you write into a task-memory artifact carries its provenance referent, and that referent travels with the claim so a later command reads it too; do not drop it at the write boundary. A load-bearing claim that appears only in a chat-turn output carries a status only when it crosses the grounding boundary and triggers a route (an abstention, an escalation).

6. Abstain as a routed continuation, never a bare refusal. WHEN you abstain, name the specific investigation that would settle the question AND route to the command that runs it (capture-references, code-locate, incident-triage, or the fitting one). A withholding that stalls the work is invalid output. Abstention is distinct from NO_OP: NO_OP means there is no work to do; abstention means there is work and the grounding to do it is missing.

7. An unfired gate is not evidence. The absence of a fired check does not mean grounding existed. Do not read silence here as a pass.

Standard output layout (required)

Produce the command output using this structure (English only):

Artifact changes

Follow ## Global output contract in WORKFLOW_OPERATING_SYSTEM.md for APPLIED / PROPOSED / SKIP rules.

Command transcript

Brief audit trail (max 4 lines; max 3 in no-op runs with NO_OP_TRACE).

Handoff

Use the adaptive ending format from WORKFLOW_OPERATING_SYSTEM.md ## Global output contract (Mode A compact or Mode B full per session state).

Definition of done (command output)

• The real run output is quoted, not asserted; a run with no shown output is reported as unverified/BLOCKED, never PASS (ADR-0048).
• Every runtime observation carries a taxonomy code and a per-criterion verdict; the gate decision (PASS | FAIL | BLOCKED) is explicit with its reason.
• The command names no specific MCP server (MCP-agnostic, DECISIONS D-1) and writes no code (a FAIL routes the fix to incident-triage or implement-slice-complement).
• A hold-until-pass loop carries the bounded-retry cap; a PASS is Layer-1 runtime evidence that does not skip Layer 2 or Layer 3.
• GODOT_RUNTIME_VERIFY.md is written in Agent mode (or PROPOSED in Ask/Plan mode per ADR-0001).
• PLAYTEST_RUNBOOK.md is written or updated with the run steps, the behaviors to exercise (including mechanic feel and fidelity the automated gate cannot judge), and where notes go; human playtest feedback is routed to pr-feedback-ingest --playtest, never absorbed into an ad-hoc review (ADR-0084).
• Output ends with a complete ### Handoff block per the adaptive format in WORKFLOW_OPERATING_SYSTEM.md ## Global output contract.
• Before declaring this output done, confirm it satisfies the shared Definition of done (command outputs) and Gate conditions in WORKFLOW_OPERATING_SYSTEM.md.

Quality bar: The verdict is only as good as the shown output. A real run with quoted errors and an honest FAIL is worth more than a confident PASS with nothing to read. Verify the runtime behavior the linter could never see, and route the fix rather than reaching for it.