kronic/Project-M
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Rewrite /dots-dev skill: Workflow-first, two-review sandwich, 6.5 pins

Browse Source 
Realigns the skill with how sessions actually run now — it predated the
Workflow tool and the ultracode two-review practice. Driven by an analysis
of the full session history + the 26-file memory corpus + skill-authoring
research, adversarially reviewed.

Key changes:
- Workflow-first orchestration: drop the dead manual "swarm (<=N agents)"
  model and the impossible 3-5-agent impl swarm; implementation is serial
  orchestrator MCP edits. New references/workflow-patterns.md (ground
  fan-out + design-review lens/critic) replaces agent-briefs.md.
- Pre-code design-review + post-impl diff-review are now first-class gated
  phases (the spine that catches what green tests + a clean Play miss).
- Size by blast-radius / netcode-heaviness, not time estimates.
- Lean SKILL.md (222 -> 131 lines, -25% KB): leans on CLAUDE.md instead of
  duplicating its MCP cheat-sheet / anti-patterns / error-recovery.
- ctx7 CLI / find-docs mechanism (the MCP verbs are gone); live-verified
  6.5-era library pins; kill the dead NetCodeTestWorld test path.
- Encode operator gates (no-time-estimates, present-forks, never-defer,
  tuning-autonomy) + the highest-recurrence MCP-edit / Workflow gotchas.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Luis Gonzalez
2026-06-18 21:19:49 -07:00
parent 6b368d1a29
commit 4ac1ae5a2e
6 changed files with 180 additions and 329 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.claude/skills/dots-dev/SKILL.md
+82 -172
View File
@@ -1,221 +1,131 @@
---
name: dots-dev
description: Run an adaptive Unity DOTS + Netcode for Entities dev session — sizes the task (trivial / small / feature), then orchestrates context7-led research, plan-mode-gated design, parallel agent swarms, DOTS-correct implementation, DOTS-aware validation, and a mandatory vault + memory update. Use when the operator types /dots-dev, "dots-dev", or opens a session on a Unity DOTS / ECS / Netcode-for-Entities project (anything like "let's work on <game>", "dots-dev: implement X", "design Y system", or other dev-session openings). Use as the default driver for any non-trivial DOTS work. Defaults all API/code research to context7. Detects a missing DOTS foundation and defers to the one-time project setup task rather than setting up inline.
description: Orchestrates a Unity DOTS (Entities) + Netcode-for-Entities dev session — grounding, design-review, plan-gated implementation, validation, and a doc bookend. Use when the operator types /dots-dev or opens DOTS/ECS/Netcode work ("let's work on <game>", "implement X", "design Y system"). The default driver for non-trivial DOTS work; defers a missing DOTS foundation to the one-time setup task.
---
# /dots-dev — Adaptive Unity DOTS / Netcode Dev Session Orchestrator
Drives an end-to-end Unity 6 development session for a **DOTS (Entities) + Netcode for Entities** game. The genre is project-specific (read it from `CLAUDE.md` + the vault); the constant is the tech stack: **data-oriented ECS, Burst, jobs, baking, server-authoritative netcode with client prediction**.
Drives a Unity 6.5 **DOTS (Entities) + Netcode for Entities** session. **`CLAUDE.md` (always-loaded) is authoritative** for conventions, build gotchas, the MCP cheat-sheet, anti-patterns, the memory layers, git rules, and the bootstrap/world model — this skill does not re-teach them; it carries the **orchestration process** CLAUDE.md lacks, and inlines only the few highest-recurrence gates it directly depends on (flagged ★). When the two disagree, `CLAUDE.md` wins.
This skill is **adaptive**: it sizes the task first, then runs only the phases that size needs. It is **reusable and path-portable** — it hardcodes no machine-specific absolute paths and works identically on macOS and Windows. It defaults every API / best-practice question to **context7** (Unity's DOTS APIs move fast). It keeps two things non-negotiable at every size: the **plan-approval gate** (`ExitPlanMode` before any state change) and the **doc + memory update** at session end.
The spine: **size by blast-radius → ground → (design-review) → plan-gate → serial implement → verify-ladder → (diff-review) → doc/memory bookend → offer commit.** Two adversarial reviews bookend the code; everything else flexes by size.
> **One-line reminder**: set max reasoning before invoking a Feature-track session. `/dots-dev` Feature track is heavy by design.
## Non-negotiables (every track)
## Core principles (read first)
1. **context7-first for anything DOTS.** Any API name, signature, pattern, or "current best practice" for Entities / Netcode / Physics / Burst / Collections is resolved through **context7** (`resolve-library-id` → `query-docs`) before it is written, and cross-checked against the official ECS Samples repo. Do not trust training-data recall for these — they change between minor versions. See [`references/context7-libraries.md`](references/context7-libraries.md).
2. **DOTS-correct, not MonoBehaviour-correct.** The conventions are ECS conventions. `ISystem`+Burst by default, `IComponentData`/buffers/enableable, baking, jobs, `EntityCommandBuffer`, determinism. The classic-Unity rules (MonoBehaviour singletons, `[SerializeField] private`, coroutines) mostly do not apply. See [`references/dots-conventions.md`](references/dots-conventions.md).
3. **Plan mode through approval.** Phases up to operator approval are read-only. Code is written only after the operator approves and `ExitPlanMode` is called. Self-police this — it is not harness-enforced.
4. **Documentation Protocol bookends.** Every non-trivial session starts with a vault/memory scan and ends with a vault/memory update. See [`references/memory-protocol.md`](references/memory-protocol.md).
5. **Cross-platform.** Discover the project root (cwd / `$CLAUDE_PROJECT_DIR`). Reference the vault repo-relative as `<repo>/Docs/Vault/`. Never write `/Users/...` or `C:\...` into code, docs, or config — committed `.mcp.json` uses `${CLAUDE_PROJECT_DIR}`.
6. **Synthesize early.** After each swarm, write a tight summary; pass only the upstream summary downstream. Never carry raw agent dumps forward.
1. **context7-first for volatile DOTS APIs ★.** Resolve Entities/Netcode/Physics/Burst/Collections API shape via the **`ctx7` CLI** (or the **`find-docs` skill**) before writing it — never training-data recall. Two-source rule: docs for the *current API*, `unity_reflect` for the *installed engine shape* (they can disagree — reflection wins for the pinned version). Pins + flow → [`references/context7-libraries.md`](references/context7-libraries.md).
2. **Plan-approval gate ★.** Phases up to approval are read-only. Code is written only after the operator approves and you call **`ExitPlanMode`**. Self-policed, not harness-enforced.
3. **Doc + memory bookend ★.** Scan the vault/memory at start; update it at end (incl. CLAUDE.md self-maintenance). Protocol → `CLAUDE.md` Memory section + [`references/memory-protocol.md`](references/memory-protocol.md).
4. **DOTS-correct.** ECS conventions per `CLAUDE.md` + [`references/dots-conventions.md`](references/dots-conventions.md). Any violation found in review is blocking.
5. **Commit only when asked.** Gated like the plan — offer + propose grouping in the final report; commit only on explicit "commit"; **never push** unless asked.
6. **Operator gates.** Present-forks-don't-auto-decide · no time/effort/deadline estimates · never-silently-defer · tuning-defaults autonomy. See *Operator gates* below.
7. **Path-portable.** Never write machine-specific absolute paths (`/Users/…`, `C:\…`) into the skill, code, docs, or `.mcp.json` — use repo-relative / `${CLAUDE_PROJECT_DIR}`.
## When to use / not use
**Use** for real DOTS work: a new system, component set, ghost/netcode surface, baking pipeline, refactor, balance pass, or design pass — on a project whose foundation already exists.
**Use** for real DOTS work on an existing foundation: a system, component set, ghost/netcode surface, baking pipeline, refactor, or balance pass.
**Do NOT use** for: one-shot questions (answer directly); mechanical edits the operator already designed (just do them); harness/tooling setup; **project setup / installing the DOTS stack** (that is the separate one-time setup task — see Pre-flight); non-DOTS work.
**Don't use** for: one-shot questions (answer directly); mechanical edits the operator already designed (just do them); harness/tooling setup; **installing the DOTS stack** (separate one-time task — Phase 0); non-DOTS work.
## Adaptive track selection (do this first)
## Size by blast-radius — do this first (NOT by time)
Size the goal, then run the matching track. Plan-approval and the doc/memory update are mandatory in **all** tracks.
Pick the track from the *surface touched*, never from a minutes estimate.
| Track | When | Phases run |
| Track | When | Path |
|---|---|---|
| **Trivial** | one-file edit, rename, a fix you already understand | inline: quick scan → do → `read_console` → one-line session-log append. No swarm, no research swarm. Still plan-gate if it mutates anything non-obvious. |
| **Small** | a single system/component with clear design | Phase 0/0b (lite) → Phase 3 context7 research (1 agent) → Phase 4 plan → **approve** → Phase 6 implement → Phase 7 validate → Phase 8 doc. |
| **Feature** | new mechanic, netcode surface, cross-cutting change, or anything >~30 min | full swarm, Phases 0–9 below. |
| **Inline** | one-file edit, rename, a fix you already understand; a pure `*Math` helper with no wire/bake change | scan → do → `read_console` clean → one-line session-log append. Plan-gate only if it mutates anything non-obvious. |
| **Small** | a single system/component, clear design, **no** ghost/prediction/RPC/relevancy/ordering/singleton-collision change | Ground (lite) → ctx7 verify → plan → **approve** → serial implement → verify L1–L2 → doc. |
| **Feature** | new mechanic, **any replicated-surface change**, cross-cutting refactor, anything on netcode-heavy surface | full spine incl. **both** reviews. The standard **ultracode** mode. |
When unsure, ask the operator with one `AskUserQuestion` ("Treat this as a quick edit or a full feature pass?") rather than guessing big.
**Blast-radius trigger** (forces Feature + the pre-code review) — touches any of: a `[GhostField]` / ghost prefab · prediction or rollback · an `IRpcCommand` / input · `GhostRelevancy` · cross-system `[UpdateBefore/After]` / group ordering · a singleton resolve where a second of that type exists. A pure `*Math.cs` helper that doesn't change wire/bake is cheap — but still run the ghost/ordering check. Unsure → one `AskUserQuestion` ("quick edit or full feature pass?"), don't guess big.
## Phase 0 — Pre-flight + intake gate
**ultracode** = the named heavy mode for Feature work: raise reasoning, run the two-review sandwich + Workflow orchestration. The project ships in numbered **Slices**; run the Phase 4 review before each netcode-heavy Slice (standing practice — see CLAUDE.md / MEMORY).
1. **Foundation check (DOTS installed?).** Read `Packages/packages-lock.json` (or `manage_packages action=list`). If `com.unity.entities` / `com.unity.netcode` are **absent**, STOP: tell the operator the project foundation isn't set up and point them to the **one-time setup task** (installs the DOTS stack, asmdefs/bootstrap, vault, memory MCPs, `CLAUDE.md`). Do **not** install packages or scaffold inline — that is out of scope for this skill.
2. **Unity ready?** Read `mcpforunity://editor/state`. If `ready_for_tools == false` or no instance is connected, surface it and ask the operator to open the Unity project. Don't spawn swarms until ready. If multiple editors are open, read `mcpforunity://instances` and `set_active_instance` on the right one.
3. **Memory stack reachable?** Quick probe: `basic-memory` (semantic recall) and `serena` (`find_symbol`). If either is down, note it and fall back (obsidian-cli + Glob/Grep). Continue — the protocol works either way. If Serena's C# LSP stalls on the Unity solution, record it and use Glob/Grep (or `claude-context` if installed) for this session.
4. **Max-reasoning nudge** (Feature track only, once): *"Feature track benefits from max reasoning — raise the slider before approving the plan."* Don't nag.
5. **Intake gate.** If scope / surface / depth / output is unclear, bundle 2–4 ambiguities into a single `AskUserQuestion` before spawning anything. If the prompt is already specific, skip the gate.
## The spine
## Phase 0b — Plan-mode entry + Documentation Protocol scan
### Phase 0 — Pre-flight
1. **Foundation check.** `com.unity.entities`/`com.unity.netcode` in `Packages/packages-lock.json`? If **absent**, STOP — point to the one-time setup task (`Docs/dots-setup-task.md`). Never install/scaffold inline.
2. **Unity ready.** `mcpforunity://editor/state` → `ready_for_tools`. Not ready / no instance → surface, ask operator to focus Unity. Multiple editors → `mcpforunity://instances` + `set_active_instance`. (An unfocused editor throttles to near-idle and breaks Burst recompiles — ask for focus on heavy build/test/Burst sessions.)
3. **Memory stack.** Probe `basic-memory` + `serena`; on failure fall back (obsidian-cli + Glob/Grep) and continue.
4. **Intake gate.** Unclear scope/surface/output → bundle 2–4 ambiguities into one `AskUserQuestion`. Specific prompt → skip.
Declare plan mode active for phases 1–5: read-only only (`Read`, `Glob`, `Grep`, MCP resource reads, `unity_reflect`, `unity_docs`, `read_console`, context7, obsidian-cli reads, `basic-memory`, serena reads). No script writes, no `manage_gameobject` mutations, no vault writes, no `Edit`/`Write`, no state-mutating `Bash`.
### Phase 1 — Ground (read-only, plan mode)
Declare plan mode: read-only tools only (`Read/Glob/Grep`, MCP resource reads, `unity_reflect`, `unity_docs`, `read_console`, ctx7/find-docs, obsidian-cli reads, `basic-memory`, serena reads). No writes.
Scan, in order (via obsidian-cli unless the Obsidian REST MCP is present):
1. `CLAUDE.md` at repo root — authoritative project conventions + game design pointers.
2. Vault map of content (`<repo>/Docs/Vault/00_Home/Home.md`) and the latest session log under `07_Sessions/<year>/`.
3. Open decision records (`_Decisions/` with `status: proposed`) + roadmap/backlog.
4. `basic-memory` semantic recall on goal keywords; native `MEMORY.md` + `memory/` for machine-local facts/feedback.
5. Task-specific vault search for the goal's nouns.
Run the **Documentation-Protocol scan** (`CLAUDE.md` → vault MoC + latest session log → open DRs/roadmap → `basic-memory` recall on the goal's nouns → native `MEMORY.md`), then ground the work along three lenses: **code/asset graph** (serena `find_symbol`/refs, else Glob/Grep on `Assets/_Project/**/*.cs` + `*.asmdef`; `manage_scene get_hierarchy`; `unity_reflect` for goal types) · **knowledge state** (vault + `basic-memory`; quote locked decisions that constrain the goal) · **live state** (`editor/state`, `read_console` baseline, installed Entities/Netcode versions for ctx7 pinning). For Feature scope, fan this out read-only via the *Ground fan-out* Workflow pattern; for Small, do it inline.
## Phase 1 — Project & doc context swarm (≤3 agents, Explore)
Synthesize a **≤400-word Project Brief**; discard raw dumps. If the goal contradicts a locked vault/`CLAUDE.md` decision, halt and surface.
- **A — Code/asset graph**: Serena `find_symbol`/references over the relevant assemblies where it works, else `Glob`/`Grep` on `Assets/_Project/**/*.cs` + `*.asmdef`; `manage_scene(action="get_hierarchy")`; `unity_reflect` for goal-relevant types.
- **B — Knowledge state**: obsidian-cli search + `basic-memory` recall for design/decisions touching the goal.
- **C — Unity live state**: `mcpforunity://editor/state`, `project/info`, `instances`; `read_console(types=["error","warning"])`; `manage_packages(action="list")` (confirm installed Entities/Netcode versions for context7 version-pinning).
### Phase 2 — ctx7 API verification
For every API/pattern the goal needs: `ctx7` CLI (or `find-docs`) pinned to the installed version, cross-checked against ECS Samples, **plus** `unity_reflect` on the installed type. Synthesize a **≤500-word Research Digest** (*steal / avoid / still-unclear* + the exact queries used). If a planned API is shown removed/renamed, halt and re-plan. **Web (`WebSearch`/`WebFetch`) is allowed ONLY for design/genre precedent — never for a DOTS API shape** (that path hallucinates). Research synthesizes-and-discards (reviews don't — Phase 4).
Synthesize into a **≤400-word Project Brief**. Discard raw dumps.
### Phase 3 — Plan → ExitPlanMode gate
Produce a **Unified Implementation Plan**: summary; exhaustive files-to-create/modify (Unity + vault, repo-relative); code skeletons (signatures only); build order with explicit "compile + `read_console`" checkpoints; the verification approach (L1→L3 ladder); **wire/bake-churn classification** (does this re-bake the subscene, change a ghost hash, or change an RPC collection hash? — re-mean bytes vs rename, per CLAUDE.md); vault updates; risks & open questions. Present it; `AskUserQuestion` on any high-stakes ambiguity. On approval, **`ExitPlanMode`** — the hard boundary. Do not start Phase 5 without it.
## Phase 2 — Goal-understanding swarm (≤3 agents)
### Phase 4 — Pre-code design review (Feature / any blast-radius trigger)
**Before writing code** for any ghost/prediction/RPC/relevancy/ordering slice, run the *Design-review (lens + adversarial-critic)* Workflow pattern over the plan + ground-truth code. Three lenses — **netcode/relevancy · determinism/prediction · reuse/scope** — then adversarial critics that try to *refute* each finding against the actual code. This catches what green tests + a clean Play miss. The synthesized, code-grounded output is a **durable Build Spec** (kept at `Docs/Vault/.../<Slice>_Build_Spec.md`) that drives the build and feeds the DR — **do not discard it.** A surfaced design flaw loops back to Phase 3. Lens prompts → [`references/workflow-patterns.md`](references/workflow-patterns.md).
Decompose the goal into a **Goal Spec**: *Deliverables* (each with a "done when"), *Assumptions* (validate at Phase 5), *Open ambiguities* (research in Phase 3 or ask). Inputs: operator's verbatim prompt + Phase 1 brief only. If the goal contradicts a locked decision in the vault/`CLAUDE.md`, halt and surface before Phase 3.
### Phase 5 — Implement (SERIAL orchestrator MCP edits — never an impl swarm)
The single stateful editor + domain-reload ordering make parallel writes unsafe — **all Unity mutations are done by the orchestrator, serially.** Parallelism is reserved for read-only research/review.
- **Edit Assets `.cs` ONLY via MCP ★** (`create_script` new / `get_sha`→`apply_text_edits` existing / `script_apply_edits` for method+anchor edits). Raw `Write`/`Edit` on `Assets/*.cs` leaves a **stale compiled assembly** (and a raw-created `.cs` gets no `.meta`/test-discovery until `refresh_unity scope=all mode=force`). `Write`/`Edit` are fine for non-asset files (vault, asmdef JSON).
- **`apply_text_edits`: one edit per call ★** (multiple non-adjacent edits can misalign), always with `precondition_sha256`.
- `read_console(types=["error"])` after **every** write; fix before moving on. `validate_script` non-trivial files first. Re-query ctx7 for any ambiguous signature. Sequence across domain reloads (a component compiles before a system/baker references it). `batch_execute` for bulk wiring.
- Further MCP-edit hazards (`script_apply_edits` can't target a `struct : ISystem`; `manage_gameobject`/`manage_prefabs` silently drop enum + Vector3 → set via `manage_components` + verify; unfocused-editor stale-Burst) → **CLAUDE.md** "MCP / editor workflow" + "Burst hazards".
- **Tuning autonomy:** within an approved plan, pick sensible defaults + expose live knobs; consult only on a real gameplay/design fork.
## Phase 3 — Research swarm (context7-FIRST)
### Phase 6 — Verify ladder (definition of done — show evidence, not "done")
NetCodeTestWorld is `internal` and unused here — do NOT prescribe it.
- **L1** — `read_console(types=["error","warning","exception"], include_stacktrace=True)` clean vs the Phase-1 baseline. Burst/source-gen failures surface here, not in a plain build — any new one is a finding.
- **L2** — EditMode tests: plain-Entities `new World → register in SimulationSystemGroup → SortSystems → Update → assert` (`run_tests mode=EditMode`). Extract pure logic to a `*Math.cs` helper and unit-test it (cover swept hit-detection with a tunnelling regression). **Plus one live Play netcode smoke** — `execute_code` boot, step ticks, assert **server == client** for the replicated surface (identify worlds by `world.Name == "ServerWorld"/"ClientWorld"`).
- **L3** (visual/asset work) — screenshot via `execute_code` `ScreenCapture` → Read PNG, **and verify asset *values*** (`shader.GetPropertyType`-guard before `GetColor`/`GetFloat`) — a dark/stylized render masks material bugs.
- **A — DOTS/Netcode implementation (context7, primary)**: for every API/pattern the goal needs, `resolve-library-id` (seeds in [`references/context7-libraries.md`](references/context7-libraries.md)) → `query-docs`. Pin to the **installed** package version. Cross-check the official ECS Samples repo for working patterns.
- **B — Determinism/netcode correctness (context7 + samples)**: prediction loop, ghost replication, command/input, world bootstrap — confirm current API shapes, not remembered ones.
- **C — Genre/design precedent (web, optional)**: only here is `WebSearch`/`WebFetch` allowed — how comparable games solve the *design* problem. Never use web for DOTS API shapes.
Loop ≤3 fix iterations per finding; if iteration 3 still fails, halt and surface "I can't resolve X without a decision."
Synthesize into a **≤500-word Research Digest**: *What to steal / What to avoid / Still unclear*, plus the exact context7 queries used (so Phase 6 can re-query precisely). If context7 reveals an API the plan depends on was removed/changed, halt and re-plan.
### Phase 7 — Post-impl diff review (Feature)
Run the *Design-review* Workflow pattern over the git diff (same three lenses). **Findings default-refuted** — each must be confirmed against the code to stand — but **flag any simplification of a reviewed design as a likely regression.** This is the phase that has caught rollback bugs, region-routing regressions, and born-correct flaws that hundreds of green tests + a clean Play all missed. Real findings loop back to Phase 5.
## Phase 4 — Design & plan swarm (≤3 agents, Plan)
- **A — Code architect**: components (`IComponentData`/`IBufferElementData`/`IEnableableComponent`), systems (`ISystem`+Burst vs `SystemBase`), jobs, system groups/order, ECB usage, asmdef impact (which assembly: shared-simulation / client / server / authoring). MUST respect [`references/dots-conventions.md`](references/dots-conventions.md).
- **B — Scene/subscene/baking architect**: authoring MonoBehaviours + `Baker<T>`, subscene layout, ghost prefabs/authoring, what bakes where, URP/Entities-Graphics interactions.
- **C — Vault doc architect**: which design docs get touched, whether a decision record is warranted, the session-log scaffold.
Synthesize a **Unified Implementation Plan**: summary; exhaustive files-to-create/modify (Unity + vault paths, repo-relative); code skeletons (signatures only); build order with explicit "compile + `read_console`" checkpoints; **DOTS-aware test approach** (EditMode? PlayMode? `NetCodeTestWorld`? thin-client? play-tick determinism check?); vault updates; risks & open questions.
## Phase 5 — Operator review (plan-mode exit)
Present the plan in full, highlight Risks & open questions. For any high-stakes ambiguity, `AskUserQuestion` with concrete options — do not proceed on assumptions. On approval, call **`ExitPlanMode`**. This is the hard boundary. **Do not start Phase 6 without explicit approval and `ExitPlanMode`.**
## Phase 6 — Implementation swarm (3–5 agents)
- **A — Script author**: `create_script` (new) / `apply_text_edits` after `get_sha` (existing). `validate_script` non-trivial files before write. `read_console(types=["error"])` after **every** write. Re-query context7 for exact signatures when the digest left anything ambiguous.
- **B — Scene/subscene/baking wirer**: authoring components, subscenes, ghost prefabs, ScriptableObject config assets, scene saves. `batch_execute` for bulk.
- **C — Asmdef/package guardian** (conditional): new assemblies or references — keep the client/server/shared/authoring split clean; verify it compiles.
- **D — Doc-skeleton author** (conditional): new vault docs get frontmatter skeletons now; bodies in Phase 8.
Parallelize only where write surfaces don't overlap. Sequence across domain reloads (a component must compile before a system or baker can reference it). Quality bar: compiles clean, no new warnings, console clean, conventions honored, Burst-compatible where it should be, no managed refs in unmanaged components, deterministic sim code, no edits during Play Mode (`manage_editor(action="play_state")` first).
## Phase 7 — Validate swarm (DOTS-aware, 3–4 agents)
- **A — Console/Burst/source-gen sweep**: `read_console(types=["error","warning","exception"], include_stacktrace=True)` vs the pre-Phase-6 baseline. Burst-compile and source-generator failures surface here, not in a plain build — any new one is a finding.
- **B — Test runner**: `run_tests(mode="EditMode")` then PlayMode if applicable, via `get_test_job(... include_failed_tests=True)`. For netcode work, a `NetCodeTestWorld` connect/tick test or a thin-client check.
- **C — Runtime/determinism spot-check**: brief Play Mode entry, step ticks, confirm predicted/interpolated ghosts behave; screenshots for visual work; verify plan-specified entities/components exist.
"Compiles clean" is necessary, not sufficient — determinism desyncs, rollback bugs, structural-change invalidation, and sync-point stalls only appear at runtime under prediction. Loop ≤3 fix iterations on a finding via a tight fix sub-swarm; if iteration 3 still fails, halt and surface a clear "I can't resolve X without a decision."
## Phase 8 — Doc + memory update (never skip)
- Session log → `<repo>/Docs/Vault/07_Sessions/<year>/<YYYY-MM-DD>_<Topic>.md` via obsidian-cli (template in the vault). Same-day collision → `_A`, `_B`, …
- Firm decisions → a decision record in `_Decisions/`, back-referenced from the session log.
- Touched design docs → edited in place (wikilink, don't duplicate into the log).
- Durable, cross-machine knowledge → a `basic-memory` note in the vault.
- Machine-local facts/feedback/working-style → native `MEMORY.md` (+ a memory file if substantive). Remember native memory does **not** sync across your Mac/Windows machines — cross-machine truth belongs in the vault.
If the operator says "skip the protocol", still write a one-line stub log so the gap is visible, and record the preference as a native feedback memory.
## Phase 9 — Final report
### Phase 8 — Doc + memory bookend (never skip)
Per `CLAUDE.md` Memory section + [`references/memory-protocol.md`](references/memory-protocol.md): session log → `Docs/Vault/07_Sessions/<year>/<YYYY-MM-DD>_<Topic>.md` (same-day → `_A`/`_B`); firm decisions → a DR in `_Decisions/` back-referenced from the log (the kept Build Spec feeds it); touched design docs edited in place (wikilink); durable cross-machine knowledge → a `basic-memory` note; machine-local facts → native `MEMORY.md`. **CLAUDE.md self-maintenance is part of this bookend** — a new gotcha updates it under its 40 KB budget (archive, don't delete); never silently defer it. "Skip the protocol" → still write a one-line stub log + record the preference as a native memory.
### Phase 9 — Final report + offer commit
```
## /dots-dev session complete — <Topic>
**Summary**: ≤4 sentences — goal, what was delivered, did it land cleanly.
**What changed**:
- Unity: <repo-relative path> — <one line> (repeat)
- Vault: <repo-relative path> — <one line> (repeat)
**Validation**: Tests <X/Y>, console clean, <netcode/determinism check result>.
**Suggested next step**: <one concrete next action, from the session log's Next-session intent>.
**Summary**: ≤4 sentences — goal, delivered, did it land cleanly.
**What changed**: Unity / Vault — <repo-relative path> — <one line> (repeat)
**Validation**: L1 console clean · L2 tests X/Y + Play smoke (server==client) · L3 <if visual>.
**Reviews**: pre-code <result> · post-impl <result>.
**Next step**: <one concrete next action from the log's Next-session intent>.
```
Then **offer to commit** and propose grouping; commit only on explicit approval. **Review first** (`git status` + `git diff --stat`); **stage explicit paths — never `git add -A`** (excludes `Library/Temp/obj/Logs/UserSettings`, `.csproj`/`.sln`, scratch screenshots/exports); keep each asset **with its `.meta`** in one commit; one commit per concern; match the repo's message style + the Co-Authored-By trailer; **never push**. Don't auto-loop into a fresh session.
Then **offer to commit** the changes (Phase 10) — propose how, but do not commit unless asked. Do not auto-loop into a fresh session.
## Workflow orchestration (the parallelism primitive)
## Phase 10 — Commit (only after explicit operator approval)
Manual "spawn ≤N agents" is dead — read-only fan-out (ground, design-review, diff-review) runs as deterministic **Workflows** with run-IDs and `resumeFromRunId`. **Implementation never fans out** (serial, Phase 5). Per-pattern lens prompts live in [`references/workflow-patterns.md`](references/workflow-patterns.md); the two cross-cutting rules:
Committing is a state change, so it is **gated like the plan**: never auto-commit. In the Phase 9 report, **offer** to commit and propose the grouping; commit only when the operator explicitly says so (e.g. "commit", "commit it", "commit for me"). If they decline or don't mention it, **leave git untouched** and note the changes are uncommitted for their review.
- **Always check the failures list ★.** A Workflow whose agents die on a session/quota limit returns **empty findings that masquerade as a clean pass.** Check failures explicitly; if any died, wait for the reset and re-run with `resumeFromRunId` — never treat an empty result as "no findings."
- **Scope subagent guardrails POSITIVELY ★.** A blanket "do NOT use tool X" in a subagent prompt gets read by the auto-approver as a *user* rule and can deny the orchestrator's own later call. Tell agents what they MAY do (read-only, these tools), not what they mustn't.
When approved:
1. **Review first.** `git status` + `git diff --stat` (and `git diff` on anything non-obvious) so the operator sees exactly what lands. **Stage explicit paths — never `git add -A` blindly.** Never commit generated/cache (`Library/`, `Temp/`, `obj/`, `Logs/`, `UserSettings/`), `.csproj`/`.sln`, or stray scratch files (debug screenshots, temp exports).
2. **Group logically** — one commit per concern, not one mega-commit: typically (a) the feature's `Assets/` code + assets, (b) tests, (c) `Docs/Vault/` + `CLAUDE.md`. Always keep a `.cs`/prefab/material/controller **with its `.meta`** in the same commit.
3. **Proper messages.** Read recent `git log --oneline` first and match the repo's style. Imperative subject (≤~72 chars) summarizing the change; a short body on the *what + why* with a back-reference to the DR / session log (e.g. "See DR-0NN"). End **every** message with the Co-Authored-By trailer the environment specifies.
4. **Branch + push policy.** Match the repo's convention (the `git log` reveals it — many solo DOTS projects commit straight to the default branch; if the history is *not* solo-commits-on-default, branch first). **Never push** unless the operator explicitly asks.
5. Report the commit hash(es) + a one-line summary of each, then stop.
Pass only the upstream summary downstream — never raw agent dumps (research discards; the design-review Build Spec is the deliberate exception — keep it).
## DOTS conventions (authoritative file)
## Operator gates
`CLAUDE.md` at the repo root is the project's source of truth; [`references/dots-conventions.md`](references/dots-conventions.md) is the reusable ECS/Netcode rule set this skill enforces. Treat any violation found in Phase 7 as a blocking finding. Quick reminders: `ISystem`+`[BurstCompile]` default; unmanaged `IComponentData`; `IEnableableComponent` over add/remove for frequently-toggled state; `Baker<T>` + subscenes for authoring; `IJobEntity`/`IJobChunk` + correct `Allocator` + disposal; `EntityCommandBuffer` for structural changes; no wall-clock in predicted sim; server-authoritative, clients send input; `[GhostField]`/`GhostAuthoringComponent` for replication. **Aspects are deprecated (Entities 1.4) — do not author new ones.** Always confirm volatile APIs via context7 at code-time.
- **Present forks, don't auto-decide.** Gameplay/design forks are the operator's call: present each with a recommendation, lock only what's approved, re-run the fork-locking ritual before each new milestone set. A Workflow that tries to auto-lock forks must be halted.
- **No time estimates.** Never frame work by minutes/effort/deadlines — the operator owns scheduling. Size by blast-radius, keep the engineering discipline.
- **Never silently defer/descope** mid-task (incl. CLAUDE.md maintenance) — do it now or explicitly ask the operator to decide.
- **Tuning-defaults autonomy** — for polish/balance, pick sensible defaults + live knobs; consult only on a real fork.
- **Aggressive-cleanup opt-in** — for an explicit cleanup session the operator may opt into max scope (deep behavior-preserving refactor, aggressive deletion, "commit for me" in logical groups on main, no push).
## MCP cheat sheet
## Boundaries
| Need | Tool / Resource | Note |
|---|---|---|
| Is Unity ready? | `mcpforunity://editor/state` | check `ready_for_tools`, `is_compiling` |
| Installed package versions | `Packages/packages-lock.json` or `manage_packages(action="list")` | pin context7 to the installed version |
| Current DOTS API/pattern | context7 `resolve-library-id` → `query-docs` | **primary** for all DOTS code |
| What scenes/objects exist | `manage_scene(action="get_hierarchy")` + `find_gameobjects` | read before mutating |
| New script | `create_script(path, contents)` → `read_console` | auto-import + compile |
| Edit script | `get_sha` → `apply_text_edits` | prefer over raw `Edit` for `.cs` in Assets |
| Validate before write | `validate_script` | catch errors pre-domain-reload |
| Bulk ops | `batch_execute(commands=[...])` | 10–100× faster |
| Run tests | `run_tests` → `get_test_job(... include_failed_tests=True)` | EditMode/PlayMode |
| Symbol nav | serena `find_symbol` / references | falls back to Glob/Grep on Unity C# issues |
| Read/search vault | obsidian-cli | REST MCP optional |
| Semantic knowledge recall | `basic-memory` | over the same vault files |
**Stop and consult** when: intake too thin · goal contradicts a locked decision · a material design choice has no clear best answer · **plan gate (always)** · 3 failed fix iterations on one finding · operator intent drifts · a destructive op outside the approved plan is implied · the DOTS foundation is missing.
## Error recovery
**May loop without consulting** (≤3): compile/Burst fix→recompile · test fail→fix→retest · vault write fail→re-read→retry.
| Symptom | Cause | Resolution |
|---|---|---|
| Tools "busy" / `is_compiling=true` | compiling | poll `editor/state` until `is_compiling==false` |
| `apply_text_edits` → `stale_file` | SHA changed | re-`get_sha`, retry |
| UnityMCP drops mid-phase | domain reload | wait ~5s, re-read `instances`, re-pin |
| `create_script` then "component not found" | reload not done | poll `is_compiling==false`, retry |
| Burst/source-gen error after a clean C# build | DOTS codegen failure | read the full console stack; fix the component/job; re-`read_console` |
| Serena `find_symbol` stalls/times out | C# LSP / .sln issue on Unity | record it; use Glob/Grep (or claude-context) this session |
| context7 version mismatch | installed pkg ≠ baked ID | re-`resolve-library-id`, prefer the installed version's docs |
| Test job timeout | slow PlayMode/netcode test | re-poll `get_test_job(wait_timeout=120)`; don't re-trigger |
**Abort without writing partial state** when: Unity unreachable >60s mid-session · project broken (asmdefs missing, won't load). If the Phase-8 vault write fails repeatedly, write a **local fallback note** + tell the operator — don't lose the session record.
### UnityMCP "not connected" — diagnostic protocol (subagents MUST follow before reporting it down)
**A bare `mcp__UnityMCP__*` error is NOT proof Unity is down** — before reporting it, run the diagnostic probe (CLAUDE.md / `mcpforunity://instances` → `editor/state` → `manage_editor telemetry_ping`; `InputValidationError` for an unknown tool = deferred schema → `ToolSearch select:` then retry). Escalate only after those fail.
A bare `mcp__UnityMCP__*` error is **not** proof the server is down. Before reporting "UnityMCP not connected", a subagent MUST run and quote results from:
1. `ReadMcpResourceTool(server="UnityMCP", uri="mcpforunity://instances")` — `instance_count >= 1` ⇒ connected (stop reporting down); `== 0` ⇒ Unity not attached (operator opens Unity).
2. `ReadMcpResourceTool(server="UnityMCP", uri="mcpforunity://editor/state")` — `ready_for_tools==false`/`is_compiling==true` ⇒ transient busy, wait 2–5s and retry.
3. `manage_editor(action="telemetry_ping")` — success ⇒ bridge alive.
4. If the call returns `InputValidationError` claiming the tool is unknown, the schema is deferred — `ToolSearch(query="select:mcp__UnityMCP__<name>")` then retry (tool-loading, not connection).
## References (load on demand)
Escalate only after 1–3 fail or step 1 returns `instance_count == 0`. Quote the literal error from each probe.
## Boundary rules
**MUST stop and consult the operator when**: intake goal too thin; the goal contradicts a locked vault/`CLAUDE.md` decision; a material design choice has no clear best answer; **Phase 5 always** (`ExitPlanMode` is the only authorized hand-off); 3 failed fix iterations on one finding; operator intent drifts from what's being built; a destructive op outside the approved plan is implied; **the DOTS foundation is missing** (defer to the setup task, don't improvise it).
**MAY loop without consulting**: compile/Burst error → fix → recompile (≤3); test failure → fix → retest (≤3); vault write failure → re-read + retry (≤2).
**MUST abort without writing partial state when**: Unity unreachable >60s mid-session; project broken (asmdefs missing, won't load); Phase 8 vault write fails repeatedly (write a local fallback note + tell the operator).
## Anti-patterns — things `/dots-dev` MUST NOT do
- Trust training-data recall for a DOTS API instead of context7.
- Author MonoBehaviour-world solutions (MonoBehaviour singletons, ScriptableObject "services" for runtime sim, `[SerializeField] private` data carriers, coroutines for sim) where ECS is correct.
- Author new `IAspect` types (deprecated in Entities 1.4).
- Store managed references in unmanaged `IComponentData`.
- Use wall-clock/`Time.deltaTime` or `System.Random` in predicted/deterministic sim.
- Do structural changes mid-job/mid-iteration instead of via `EntityCommandBuffer`.
- Edit `.meta` files directly, or read/write `Library/`, `Temp/`, `obj/`, `Logs/`, `UserSettings/`.
- Edit/commit `*.csproj` / `*.sln` (Unity regenerates them). Edit `.asmdef` / `manifest.json` instead.
- Edit during Play Mode (check `play_state` first).
- Write machine-specific absolute paths into the skill, code, docs, or `.mcp.json` (use `<repo>` / `${CLAUDE_PROJECT_DIR}`).
- Install the DOTS stack / scaffold the project inline (that's the separate setup task).
- Skip plan mode, start Phase 6 without `ExitPlanMode`, or skip the Phase 8 doc/memory update.
- **Commit or push without explicit operator approval** — committing is gated like the plan (Phase 10): offer it in the report, commit only when the operator says so, group logically with proper messages, and **never push** unless asked. Run more than one `/dots-dev` in parallel.
## References
- [`references/dots-conventions.md`](references/dots-conventions.md) — ECS/Netcode conventions this skill enforces.
- [`references/context7-libraries.md`](references/context7-libraries.md) — seed library IDs + query playbook.
- [`references/agent-briefs.md`](references/agent-briefs.md) — per-phase swarm brief templates.
- [`references/memory-protocol.md`](references/memory-protocol.md) — vault / basic-memory / serena / native-memory usage + bookends.
- [`references/dots-conventions.md`](references/dots-conventions.md) — the reusable ECS/Netcode rule set this skill enforces (defers to `CLAUDE.md`).
- [`references/context7-libraries.md`](references/context7-libraries.md) — `ctx7`/`find-docs` flow + 6.5-era pinned library IDs.
- [`references/workflow-patterns.md`](references/workflow-patterns.md) — the named Workflow patterns (ground fan-out · design-review lens + adversarial-critic).
- [`references/memory-protocol.md`](references/memory-protocol.md) — the four memory layers + bookend detail.