| name | omg-cook |
| description | Implement features end-to-end: plan, code, test, review via registry agents. Use for 'implement X', 'build Y feature', 'add Z functionality'. Handles full workflow. |
Codex Port Notice
This skill was ported from upstream reference material. Interpret command names, paths, and agent-routing guidance as Codex/Oh My Game Kit equivalents. Prefer active Codex tools and local project instructions over Codex-specific mechanics when they conflict.
Oh My Game Kit Cook -- Feature Implementation
End-to-end feature implementation. Routes to the registered implementer agent for the current kit.
Principles: YAGNI, KISS, DRY | Token efficiency | Concise reports
Tool guard — AskUserQuestion is deferred
AskUserQuestion is a deferred tool: its name appears in the deferred-tools system-reminder but its schema is NOT loaded at session start. Direct invocation fails with InputValidationError.
Operational pre-step (mandatory before drafting any structured multi-option question):
- Verify
AskUserQuestion is in the loaded tool list. If not, run:
ToolSearch(query="select:AskUserQuestion", max_results=1)
- THEN draft and invoke the tool with batched options.
Failure mode this guard prevents: assistant remembers the rule, drafts the question correctly in its head, then because the tool isn't loaded, falls back to "I'll just write the options as prose, and call the tool next time." Drafting prose bullets first is a violation — see rules/always-ask-on-unresolved.md "Forbidden prose" table.
Decision tree — which path do I take?
Pick by intent; keep loading minimal.
| Intent | Path |
|---|
| Ship a feature end-to-end with research + tests + review | Default --interactive workflow below |
| Plan exists already; execute it | Pass plan path; mode auto-detects to code |
| Skip research; you trust the spec | --fast (still requires plan step) |
| Multiple unrelated features at once | --parallel (sub-agents on disjoint files) |
| Test-driven: tests first, then code | --tdd (incompatible with --parallel / --no-test) |
Usage
omg-cook <natural language task OR plan path>
IMPORTANT: If no flag is provided, the skill uses interactive mode by default.
Optional flags: --interactive (default) | --fast (skip research) | --parallel (multi-agent) | --no-test | --auto (auto-approve) | --tdd (test-driven: write tests first, implement, verify)
Agent Routing
Follow protocol: .agents/skills/omg-cook/references/routing-protocol.md — role: implementer
Skill Activation
Follow protocol: .agents/skills/omg-cook/references/activation-protocol.md
Do NOT write implementation code until a plan exists and has been reviewed.
Exception: `--fast` mode skips research but still requires a plan step.
User override: If user explicitly says "just code it" or "skip planning", respect their instruction.
Smart Intent Detection
| Input Pattern | Detected Mode |
|---|
Path to plan.md or phase-*.md | code — execute existing plan |
| Contains "fast", "quick" | fast — skip research |
| Contains "trust me", "auto" | auto — auto-approve all steps |
| Lists 3+ features OR "parallel" | parallel — multi-agent |
| Contains "no test", "skip test" | no-test — skip testing |
| Contains "tdd", "test first", "test-driven" | tdd — write tests before implementation |
| Default | interactive — full workflow |
Full detection logic: references/intent-detection.md
Workflow
[Intent Detection] -> [Research?] -> [Review] -> [Plan] -> [Review] -> [Implement] -> [Review] -> [Test?] -> [Review] -> [Finalize]
| Mode | Research | Testing | Review Gates |
|---|
| interactive | yes | yes | User approval at each step |
| auto | yes | yes | Auto if score>=9.5 |
| fast | no | yes | User approval at each step |
| parallel | optional | yes | User approval at each step |
| no-test | yes | no | User approval at each step |
| code | no | yes | User approval per plan |
| tdd | yes | yes (3.T/3.I/3.V) | User approval at each step |
Full step definitions: references/workflow-steps.md
TDD sub-step details: references/workflow-steps.md → ## --tdd Flag Behavior
Review processes: references/review-cycle.md
Guards
--tdd + --parallel: REFUSE with error: "TDD requires strict ordering (tests → implement → verify); parallel execution cannot preserve this. Use --tdd alone, or --parallel without --tdd. For a fast sequential path, consider --tdd --fast."--tdd + --no-test: REFUSE with error: "TDD mode inherently requires the test suite; --no-test is contradictory. Remove one of the flags."--tdd + --parallel is unsupported and will error. Do not attempt to combine them.--auto + --interactive: existing guard, unchanged.
Required Subagents (MANDATORY)
Testing, Review, and Finalize phases MUST use Task tool to spawn subagents — DO NOT inline these steps.
Full subagent table and injection protocol: references/subagent-patterns.md
Finalize (never skip): omg-project-manager → plan sync-back | omg-docs-manager → update ./docs | omg-git-manager → commit offer
- When you discover a non-obvious gotcha while implementing, emit a
[omg-lesson kit="..." skill="..." fragment="..." reason="..."] marker
in your final message. The lesson-collector hook queues it for
follow-up sync-back automatically.
Blocking Gates (Non-Auto Mode)
Human review required at: Post-Research, Post-Plan, Post-Implementation, Post-Testing (100% pass required).
Always enforced: 100% test pass (unless no-test), code review score>=9.5 for auto-approve.
When implementation goes wrong — recovery via rollback
If an implementation corrupts ~/.agents/ state (bad merge, mis-applied
prefix, broken hooks), use the H7 rollback command rather than a manual
clean-up:
omg rollback --kit <name> --to-snapshot pre-<previous-version>
- Snapshots live at
~/.agents/.omg-snapshots/<kit>/pre-<version>/.
Cap is 5 most-recent per kit; older ones soft-move to
~/.agents/.omg-trash/<kit>/.
--to-snapshot MUST literally start with pre- (e.g. pre-2.4.1).
Bare versions are rejected.--yes is currently a no-op as of cli@v4.14.0 — does not bypass
anything. Future-proof only.- Restore is a non-destructive copy: files added by the bad implementation
are NOT removed. Run
omg-doctor afterwards to spot stragglers.
Critical caveat (cli@v4.14.0): the install/update pipeline does not
yet auto-create snapshots. If omg rollback reports
snapshot '<kit>/pre-<version>' does not exist, fall back to
omg install --reset (the sanctioned destructive path; takes its own
~/.agents-backup-{ISO-ts}/ first). NEVER rm -rf ~/.agents/ —
the validate-no-raw-rm-codex.cjs gate forbids it and you'll lose
.omg-snapshots/, .omg-trash/, and any user-customized files.
For per-file/per-step recovery, prefer git restore on the project
side; the H7 rollback is for ~/.agents/ state, not project source.
Full reference: .agents/skills/omg-kit/references/cli-commands.md →
omg rollback.
Environment Variables
OMG resolves env vars in priority order — never hardcode values. Details: references/env-hierarchy.md
References
references/intent-detection.md — detection rules and routing logicreferences/workflow-steps.md — detailed step definitions for all modesreferences/review-cycle.md — interactive and auto review processesreferences/subagent-patterns.md — subagent invocation patternsreferences/env-hierarchy.md — .env resolution hierarchyreferences/god-prefab-extraction-risk.md — plan-phase detection for god-prefab moves to Addressables; required audit during Step 2 when removing prefabs from scenes
Contribution Scoring
If Finalize opens a PR, invoke omg-contribution-score with type=sync-back-pr + PR URL/title/body. Fire-and-forget; SSOT gates non-OMG repos. See .agents/skills/omg-contribution-score/SKILL.md.
Sub-Agent Fork Hygiene
Sub-agent forking: see .agents/skills/omg-architecture/references/fork-hygiene.md.
