Tweak

Fast path for small Godot 4.x changes: a bugfix, a tuning/value adjustment, a small script or scene edit that fits 1-3 files. Everything heavier belongs to /game-ship (or /game-debug for hard bugs) — the gate that decides is shared/TWEAK-DISCIPLINE.md (size gate, backlog guard, registration policy, never-do list). No TaskCreate phase tracking — a tweak run is minutes of work with no compaction risk; ceremony is what this skill exists to avoid. Skill file stays English; user-facing output follows CLAUDE.md § User Preferences → Language:.

Tweak configuration (per shared/TWEAK-DISCIPLINE.md):

• verify: gdlint + scoped GUT tests per PHASE 3 below
• escalation ship target: /game-ship
• escalation debug target: /game-debug (standalone skill)

PHASE 0 — Pre-flight, size gate & backlog guard

1. Repo: resolve $REPO to the main worktree (per shared/SYNC.md Worktree-aware Path Resolution). .project/ absent → degrade gracefully: skip the card lookup, guard, and learnings silently, keep the rest (the code change is the value; do not scaffold).
2. Mode + description: .project/ present → load once: node ~/.claude/scripts/backlog-load.js "$REPO" guard-items (the game-pipeline stage field included; reused by step 6's guard — one load, not two). The invocation argument matches a live POLISH card — exact name, or an unambiguous ≥2-shared-token match — per shared/TWEAK-DISCIPLINE.md § Card pickup path 1?
   • Match → card mode. Description = the card's description. Run the dependency check now (one AskUserQuestion only if the card has an open, unshipped dependency).
   • No match, or .project/ absent → free-text mode. Description from the invocation argument; if empty, ask one short question.
3. Branch guard per shared/TWEAK-DISCIPLINE.md § Branch guard: not on the default branch (or detached HEAD) → warn + ask before proceeding.
4. Baseline per shared/SCOPED-COMMIT.md § 1: git status --porcelain | sort > .project/session/pre-tweak-status.txt (skip when .project/ is absent).

5. Gate + guard: Read shared/TWEAK-DISCIPLINE.md; run § Size gate on the projected scope (game surface examples: new scene, autoload, signal contract, input action). § Backlog guard: card mode → skip (the card already names the scope). Free-text mode → run § Backlog guard as before — the guard-items load includes the game-pipeline stage field — then § Card pickup path 2 (the POLISH-only mini-guard) against the same guard-items load from step 2.

   Todo: if any gate criterion fires or the guard demands escalation → Read .claude/skills/game-tweak/references/escalate.md and follow it — never continue silently.

6. Slug: card mode → the card's own kebab name (no re-derivation). Free-text mode → derive a kebab-case slug from the description (commit scope + learnings feature key). Print the status: Gate: ✓ tweak-sized · Guard: ✓ no card overlap (or the warn/advisory lines; card mode prints Card: {name} instead of the guard line).

PHASE 1 — Locate & context

1. Locate the change with minimal reads (Grep → targeted Read; .tscn files: read only the relevant node sections). The files found here feed the size-gate re-check and the learnings load below.

   Pre-seeded from a ship de-escalation handoff (invocation carries a files[] + acceptance[] pair, per shared/TWEAK-DISCIPLINE.md § De-escalation gate (a)) → skip the Grep/Read locate itself; use the passed files[] as the located set directly. The size-gate re-check still runs against that set — the handoff is not an exemption from it. Hold the passed acceptance[] in memory and use it as PHASE 3's verify checklist instead of relying on whatever coverage happens to already exist.

   Todo (card mode): locate shows the described defect is already resolved on main — a later commit fixed it, or it never applied → stale card. Do not invent a change to justify the card. Skip PHASE 2, PHASE 3, and PHASE 4 step 1 entirely (including step 2's learnings load); go straight to the PHASE 4 card-completion write per shared/TWEAK-DISCIPLINE.md § Card pickup → Stale card: shippedSha = the resolving commit (git log -- <file>), or HEAD if none pins it; summary names the card stale; commit nothing.

   Todo (card mode): locate/analysis (or an explicit user call mid-run) shows the card's whole reason to exist is superseded by a different, wider card — not fixed, just made moot → obsolete/superseded card. Confirm with one AskUserQuestion naming the superseding card before touching anything. On confirmation: skip PHASE 2, PHASE 3, and PHASE 4 step 1 entirely; go straight to the § Card pickup → Obsolete/superseded card cancellation write. On decline: continue the tweak as originally scoped.

2. Learnings — mandatory, not gated on tweak size. Run exactly this, after locate so --paths carries the real file anchors (see shared/LEARNINGS-LOAD.md for the full parameter/relevance model):

   node ~/.claude/scripts/learnings-search.js "$REPO" load \
     --feature "{slug}" --scopes component --pitfall-prefix true \
     --paths "{located files, comma-separated, repo-relative}"

   Any non-zero exit is a failed step, not a zero-match result — fix the invocation and re-run. Only exit 0 with empty stdout is a genuine zero. Include the printed block verbatim when non-empty, skip silently otherwise.

PHASE 2 — Implement

1. Conditional plan mode (shared/PLAN-MODE.md § Difficulty escalation): call EnterPlanMode only when the root cause is still unclear after first evidence, or the change spans > 2 files. Enter → design → write the decision to the plan file → ExitPlanMode → execute. Skip entry when plan mode is already active. The typical tweak never enters.
2. Bugfix-shaped tweaks follow shared/DEBUG-LADDER.md tier 1/2: hypothesis before edit, evidence before a second attempt — never guess-and-check.
3. Mid-flight re-check: the moment actual scope exceeds the size gate (a 4th file, a discovered new surface) → stop and Read references/escalate.md.

PHASE 3 — Verify light

A tweak that changes no runnable code — only docs or gitignored .project/ state (e.g. recording a known issue as a learning) — or a stale or obsolete/superseded card with nothing to edit (see PHASE 1) — has nothing to verify and nothing to commit: skip PHASE 3 and PHASE 4 step 1, say so in the report (no commit sha line), and go straight to the card-completion (or cancellation) + learning writes. Everything below assumes a code change.

Scoped to the touched scripts/scenes — never the full suite unless it is genuinely fast:

• Lint: gdlint on changed .gd files (gdformat already runs via the format-on-save hook).
• Tests: run the GUT test file(s) covering the touched scripts: godot --headless --path . -s addons/gut/gut_cmdln.gd -gtest={test-file} -gexit (resolve the Godot executable per .claude/paths.local.yaml → CLAUDE_GODOT_EXECUTABLE).
• No covering test + behavior change: ask the user for one quick scene-run confirmation instead of writing a new test file (a new test file is size-gate criterion 3).

New failures vs the baseline → fix within the current tier's discipline; unfixable within tweak scope → Read references/escalate.md.

PHASE 4 — Wrap-up

1. Scoped commit per shared/SCOPED-COMMIT.md. Deltas: baseline pre-tweak-status.txt; OVERLAP policy auto-include (the fix is the point); fallback: ask which files belong to the tweak; message {fix|refactor|perf|style|test|docs|chore}({slug}): {summary} — never feat (net-new capability is an escalation criterion by definition), test when the only change is added or expanded test coverage, docs for a docstring/comment/markdown-only edit; cleanup: remove the baseline file.

2. Card-mode completion (skip entirely in free-text mode): per shared/TWEAK-DISCIPLINE.md § Card pickup completion write — flip the card shipped: true + shippedAt + shippedSha + summary (this tweak's one-line outcome), remove a board-set transition field if present (it was only ever the board's queue-marker, never a lifecycle state — see § Never), and move it from backlog.json#features[] to .project/archive/backlog-archive.json#archived[]. project.json persists no features list — the dashboard derives features from backlog + archive. Then re-read backlog.json and confirm the card left features[] before reporting shipped — a running board app (serve-backlog.js) re-serializes that file from its own in-memory store and can silently revert an external write; on a revert, re-apply and re-verify.

   Obsolete/superseded card instead (PHASE 1's confirmed obsolete branch): run the § Card pickup → Obsolete/superseded card cancellation write instead — in place within features[], flip status: "CANCELLED", add cancelledReason: "superseded by {card}: {one-line why}" and cancelledAt, remove transition. The card stays in features[] (never moves to the archive — that move is shipped-only). Same board-app revert guard: re-read backlog.json and confirm status: "CANCELLED" survived before reporting; re-apply on a revert.

3. Optional learning (0-1): only for a bugfix whose root cause has value beyond this spot (filter per shared/LEARNING-WRITE.md § Writer Append Protocol) — append via learnings-write.js append with type: "pitfall", source: "extracted", 0-3 tags (game vocabulary: godot, gdscript, scene, game-loop), then run the Consolidation Gate once (LEARNING-WRITE.md § Consolidation Gate). Skip both silently otherwise. No state auto-push (TWEAK-DISCIPLINE § Registration policy).
4. Report (compact prose, no rigid table): what changed with file:line refs, checks run, commit sha, a Guard: line repeating any card overlap (card mode prints Card: {name} → shipped, or Card: {name} → cancelled (superseded by {card}) for the obsolete/superseded outcome), a Learning: line when one was written, and Escalation overridden: {criterion} when applicable. Add Next steps: /game-ship {card} only when the guard flagged a TODO card (free-text mode only — card mode is already terminal). A tweak is terminal — no next-step offer otherwise.