Skip to main content

create-control-manifest

Flat must-do/never-do rules sheet per system and layer, extracted from Accepted ADRs. ADRs explain why; this is actionable.

Quellinformationen

Repository
Donchitos/Claude-Code-Game-Studios
Letzte Quellaktivität
29. September 2026 um 11:32
Erkannte Sprache von SKILL.md
Englisch
Sterne
26.158
Forks
3.702

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
create-control-manifest
description
Flat must-do/never-do rules sheet per system and layer, extracted from Accepted ADRs. ADRs explain why; this is actionable.
argument-hint
[update — regenerate from current ADRs] [--review full|lean|solo]
user-invocable
true
allowed-tools
Read, Glob, Grep, Write, Agent, AskUserQuestion, Bash(bash "*/.claude/skills/create-control-manifest/../../hooks/yaml-helper.sh" resolve_config *)
model
sonnet
!`bash "${CLAUDE_SKILL_DIR}/../../hooks/yaml-helper.sh" resolve_config --keys review_mode,automation` Resolved above — use as-is; `--review` overrides `review_mode` for this run. No block → defaults in `.claude/docs/config-resolution.md`. # Create Control Manifest The Control Manifest is a flat, actionable rules sheet for programmers. It answers "what do I do?" and "what must I never do?" — organized by architectural layer, extracted from all Accepted ADRs, technical preferences, and engine reference docs. Where ADRs explain *why*, the manifest tells you *what*. **Output:** `docs/architecture/control-manifest.md` **When to run:** After `/architecture-review` passes and ADRs are in Accepted status. Re-run whenever new ADRs are accepted or existing ADRs are revised. --- Every `AskUserQuestion` call follows `.claude/docs/automation-modes.md` (collaborative asks always · guided major-only · autonomous logs and proceeds; `automation_always_ask` categories always prompt). ## 1. Load All Inputs ### ADRs **Establish the denominator first.** Glob `docs/architecture/adr-*.md`. Call the count **N**. If N is 0: "No ADRs found — run `/architecture-decision` before building a control manifest." Stop. **Resolve status without reading the ADRs** — the filter must precede the read, not follow it: ``` Grep pattern="^## Status" glob="docs/architecture/adr-*.md" output_mode="content" -A 3 ``` Interpret against N: | Result | Meaning | Action | |---|---|---| | **N matches, some read `Accepted`** | Normal. | The Accepted set is **A**; proceed with A. Name every ADR left out, with its status, in the Phase 4 preview — the user approves a manifest knowing what it does not cover. | | **N matches, none `Accepted`** | Genuinely no Accepted ADRs. | "[N] ADRs found, none Accepted. A manifest built from Proposed ADRs would encode decisions that may still change." Ask whether to proceed with Proposed or stop. **Do not silently emit an empty manifest.** | | **0 matches, N > 0** | **Malformed ADRs — not an empty Accepted set.** `## Status` is BLOCKING-if-missing. | "[N] ADRs found, none has a `## Status` section — acceptance cannot be determined. Run `/architecture-decision retrofit [file]` on each." **Stop.** Do not treat all ADRs as Accepted; do not emit a manifest. | - Note the ADR number and title for every rule sourced. ### Project Config - Read `naming.*` and `performance.*` from `project.yaml`; for any key absent or empty, fall back to `.claude/docs/technical-preferences.md` - Read approved libraries/addons and forbidden patterns from `.claude/docs/technical-preferences.md` (not migrated to project.yaml) ### Engine Reference - Read `docs/engine-reference/[engine]/VERSION.md` for engine + version - Read `docs/engine-reference/[engine]/deprecated-apis.md` — these become forbidden API entries - Read `docs/engine-reference/[engine]/current-best-practices.md` if it exists ### Existing Artifacts - Glob `docs/architecture/control-manifest.md` (present → this run is a regeneration; Read it before the Phase 5 write, which overwrites it) and `production/epics/*/EPIC.md` (present → epics exist). Phase 6 picks its next step from both. Report: "Loaded [N] Accepted ADRs, engine: [name + version]." --- ## 2. Extract Rules from Each ADR Read **only these four sections** per Accepted ADR — not the whole file. Context, Consequences, Migration Plan and Validation Criteria explain *why* a decision was made; the manifest records *what to do*, so they are not needed here: ``` Grep pattern="^## (Decision|Alternatives Considered|Performance Implications|Engine Compatibility)" glob="docs/architecture/adr-*.md" output_mode="content" -A 30 ``` Filter the results to the Accepted set **A** resolved above. If a section is absent for a given ADR, note it per ADR and continue; if a section is absent across **all** of A, report "No Accepted ADR contains a `[section]` section — the manifest's [category] rules will be empty. Verify this is intended." Escalate to a full read of one ADR only when its scanned sections cross-reference material outside them (e.g. a Decision that says "subject to the constraints in Context"). For each Accepted ADR, extract: ### Required Patterns (from the `## Decision` section) - Every "must", "should", "required to", "always" statement in the Decision body — including any `### Implementation Guidelines` sub-heading when the ADR has one (newer ADRs emit it; older ones state mandates directly in `## Decision`). Never scan for `### Implementation Guidelines` alone — it is absent from many skill-authored ADRs, and scanning for it yields an empty Required Patterns section on a manifest that should have been full. - Every specific pattern or approach mandated ### Forbidden Approaches (from "Alternatives Considered" sections) - Every alternative that was explicitly rejected — *why* it was rejected becomes the rule ("never use X because Y") - Any anti-patterns explicitly called out ### Performance Guardrails (from "Performance Implications" section) - Budget constraints: "max N ms per frame for this system" - Memory limits: "this system must not exceed N MB" ### Engine API Constraints (from "Engine Compatibility" section) - Post-cutoff APIs that require verification - Verified behaviours that differ from default LLM assumptions - API fields or methods that behave differently in the pinned engine version ### Layer Classification Classify each rule by the architectural layer of the system it governs: - **Foundation**: Scene management, event architecture, save/load, engine init - **Core**: Core gameplay loops, main player systems, physics/collision - **Feature**: Secondary systems, secondary mechanics, AI - **Presentation**: Rendering, audio, UI, VFX, shaders If an ADR spans multiple layers, duplicate the rule into each relevant layer. --- ## 3. Add Global Rules Combine rules that apply to all layers. Each carries its source into the manifest, as a layer rule does — the `project.yaml` key, `technical-preferences.md`, or the engine-reference file: ### From project config (`project.yaml`, else `technical-preferences.md`): - Naming conventions — `naming.*`: classes, variables, signals/events, files, constants - Performance budgets — `performance.*`: target framerate, frame budget, draw call limits, memory ceiling ### From deprecated-apis.md: - Deprecated APIs → Forbidden API entries, **filtered for relevance to this project**. Do not copy the deprecation table wholesale. > Each entry must apply to what this project actually builds. `deprecated-apis.md` > mixes dimensions and subsystems: its `GodotPhysics3D → Jolt Physics 3D` row is > **3D-only**, so emitting it unqualified gives a 2D project a global rule about > a physics engine it never uses. If you cannot tell whether an entry applies — > 2D vs 3D, a module the project does not include — either scope the rule > ("when using 3D physics: ...") or omit it and note it as unresolved. A manifest > of rules that do not apply is one nobody reads, and `/create-stories` consumes > this file. ### From current-best-practices.md (if available): - Engine-recommended patterns → Required entries ### From technical-preferences.md forbidden patterns: - Copy any "Forbidden Patterns" entries directly --- ## 4. Present Rules Summary Before Writing Before writing the manifest, present a summary to the user: ``` ## Control Manifest Preview Engine: [name + version] ADRs covered: [list ADR numbers] ADRs excluded: [ADR-NNNN ([status]), … — every ADR not in the Accepted set, or "None"] Total rules extracted: - Foundation layer: [N] required, [M] forbidden, [P] guardrails - Core layer: [N] required, [M] forbidden, [P] guardrails - Feature layer: ... - Presentation layer: ... - Global: [N] naming conventions, [M] forbidden APIs, [P] approved libraries ``` Use `AskUserQuestion`: - Prompt: "Does this rule summary look complete?" - Options: - `[A] Yes — looks good, run the director review and write the manifest` - `[B] Add rules — I have additional rules to include before writing` - `[C] Remove rules — some extracted rules should be dropped` - `[D] Stop here — I need to review the ADRs first` --- ## 4b. Director Gate — Technical Review **Review mode check** — apply before spawning TD-MANIFEST: - `solo` → skip. Note: "TD-MANIFEST skipped — Solo mode." Proceed to Phase 5. - `lean` → skip. Note: "TD-MANIFEST skipped — Lean mode." Proceed to Phase 5. - `full` → spawn as normal. Spawn `technical-director` via `Agent` using gate **TD-MANIFEST** (`.claude/docs/director-gates/td-manifest.md`). Pass: the Control Manifest Preview from Phase 4 (rule counts per layer, full extracted rule list), the list of ADRs covered, engine version, and any rules sourced from technical-preferences.md or engine reference docs. The technical-director reviews whether: - All mandatory ADR patterns are captured and accurately stated - Forbidden approaches are complete and correctly attributed - No rules were added that lack a source ADR or preference document - Performance guardrails are consistent with the ADR constraints Apply the verdict: - **APPROVE** → proceed to Phase 5 - **CONCERNS** → surface via `AskUserQuestion` with options: `Revise flagged rules` / `Accept and proceed` / `Discuss further` - **REJECT** → do not write the manifest; fix the flagged rules and re-present the summary - **NOT ASSESSED** [missing input] → not an approval (`.claude/docs/director-gates.md`): name what was missing, then supply it and re-run the gate — or, if the user chooses to go on without it, continue to Phase 5 and state `TD-MANIFEST: NOT ASSESSED — [input]` in the output and the final Verdict line --- ## 5. Write the Control Manifest Use `AskUserQuestion`: - Prompt: "May I write the Control Manifest?" - Options: - `[A] Yes — write to docs/architecture/control-manifest.md` - `[B] Show me the full draft first, then ask again` - `[C] Not yet — I want to make more changes` Format: ```markdown # Control Manifest > **Engine**: [name + version] > **Last Updated**: [date] > **Manifest Version**: [date] > **ADRs Covered**: [ADR-NNNN, ADR-MMMM, ...] > **Status**: [Active — regenerate with `/create-control-manifest update` when ADRs change] `Manifest Version` is the date this manifest was generated. Story files embed this date when created. `/story-readiness` compares a story's embedded version to this field to detect stories written against stale rules. Always matches `Last Updated` — they are the same date, serving different consumers. This manifest is a programmer's quick-reference extracted from all Accepted ADRs, technical preferences, and engine reference docs. For the reasoning behind each rule, see the referenced ADR. --- ## Foundation Layer Rules *Applies to: scene management, event architecture, save/load, engine initialisation* ### Required Patterns - **[rule]** — source: [ADR-NNNN] - **[rule]** — source: [ADR-NNNN] ### Forbidden Approaches - **Never [anti-pattern]** — [brief reason] — source: [ADR-NNNN] ### Performance Guardrails - **[system]**: max [N]ms/frame — source: [ADR-NNNN] --- ## Core Layer Rules *Applies to: core gameplay loop, main player systems, physics, collision* ### Required Patterns ... ### Forbidden Approaches ... ### Performance Guardrails ... --- ## Feature Layer Rules *Applies to: secondary mechanics, AI systems, secondary features* ### Required Patterns ... ### Forbidden Approaches ... --- ## Presentation Layer Rules *Applies to: rendering, audio, UI, VFX, shaders, animations* ### Required Patterns ... ### Forbidden Approaches ... --- ## Global Rules (All Layers) ### Naming Conventions | Element | Convention | Example | Source | |---------|-----------|---------|--------| | Classes | [from naming.* in project.yaml, else technical-preferences.md] | [example] | [`naming.classes` in project.yaml, or technical-preferences.md] | | Variables | [from naming.* in project.yaml, else technical-preferences.md] | [example] | [`naming.variables` in project.yaml, or technical-preferences.md] | | Signals/Events | [from naming.* in project.yaml, else technical-preferences.md] | [example] | [`naming.signals` in project.yaml, or technical-preferences.md] | | Files | [from naming.* in project.yaml, else technical-preferences.md] | [example] | [`naming.files` in project.yaml, or technical-preferences.md] | | Constants | [from naming.* in project.yaml, else technical-preferences.md] | [example] | [`naming.constants` in project.yaml, or technical-preferences.md] | ### Performance Budgets | Target | Value | Source | |--------|-------|--------| | Framerate | [from performance.* in project.yaml, else technical-preferences.md] | [`performance.target_framerate` in project.yaml, or technical-preferences.md] | | Frame budget | [from performance.* in project.yaml, else technical-preferences.md] | [`performance.frame_budget_ms` in project.yaml, or technical-preferences.md] | | Draw calls | [from performance.* in project.yaml, else technical-preferences.md] | [`performance.draw_call_limit` in project.yaml, or technical-preferences.md] | | Memory ceiling | [from performance.* in project.yaml, else technical-preferences.md] | [`performance.memory_ceiling_mb` in project.yaml, or technical-preferences.md] | ### Approved Libraries / Addons - [library] — approved for [purpose] — source: technical-preferences.md ### Forbidden APIs ([engine version]) These APIs are deprecated or unverified for [engine + version]: - `[api name]` — deprecated since [version] / unverified post-cutoff - Source: `docs/engine-reference/[engine]/deprecated-apis.md` ### Cross-Cutting Constraints - [constraint that applies everywhere, regardless of layer] — source: [ADR-NNNN, preference key or engine-reference file] ``` --- ## 6. Suggest Next Steps After writing the manifest: - If epics/stories don't exist yet: "Run `/create-epics layer: foundation` then `/create-stories [epic-slug]` — programmers can now use this manifest when writing story implementation notes." - If this is a regeneration (manifest already existed): "Updated. Recommend notifying the team of changed rules — especially any new Forbidden entries." --- ## Collaborative Protocol **Applies in `collaborative` mode (the default).** For `guided` and `autonomous` modes, see `.claude/docs/automation-modes.md` — the rules below describe what collaborative mode requires, not universal behavior. 1. **Load silently** — read all inputs before presenting anything
Auf GitHub ansehen
Diese SKILL.md ist sehr gross, daher zeigt SkillsMP hier nur den ersten Abschnitt. Auf GitHub ansehen