Name: speckit
Author: wewececetw

Prompt file imported from wewececetw/task-router (.codex/prompts/speckit.md). Copyright stays with the author.

Spec Kit With Codex

Use GitHub Spec Kit as the source of truth for SDD workflow. .specify/ scripts/templates drive the workflow; oMLX only accelerates selected simple outputs. Codex should match the legacy Claude Code routing rules for shared Task Router categories; only the guidance format differs.

If .specify/ is missing, initialize Codex support first:

specify init --here --ai codex --offline

For existing Spec Kit projects, prefer repo-local scripts:

.specify/scripts/bash/create-new-feature.sh
.specify/scripts/bash/setup-plan.sh
.specify/scripts/bash/check-prerequisites.sh
.specify/scripts/bash/update-agent-context.sh codex

Responsibility Split

Codex handles constitution, specify, clarify, plan, review, security-sensitive logic, and ambiguous product decisions.
The local model handles tasks, checklist, analyze, summaries, and repetitive transformations through ./scripts/call-omlx.sh.
Simple implementation tasks such as boilerplate, CRUD, config, docstrings, migration drafts, and test stubs may call oMLX.
Complex implementation tasks such as auth, security, core logic, concurrency, and data-loss risk stay with Codex.
If a preset validator fails, Codex fixes the output directly.

Commands

Create or update Codex agent context: .specify/scripts/bash/update-agent-context.sh codex
Generate with local preset: ./scripts/call-omlx.sh "$(cat plan.md)" --preset speckit-tasks
Checklist with local preset: ./scripts/call-omlx.sh "$(cat spec.md plan.md)" --preset speckit-checklist
Analyze with local preset: ./scripts/call-omlx.sh "$(cat spec.md plan.md tasks.md)" --preset speckit-analyze

Quality Gates

Specs must be testable, technology-agnostic, and limited to at most three explicit clarification markers.
Plans must identify language, dependencies, storage, project type, and constitution checks.
Tasks must use T001, T002, and so on; mark parallel work with [P] only when files and dependencies are independent.
Implementation must preserve existing user changes and keep patches scoped.

Install speckit and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

# Spec Kit With Codex Use GitHub Spec Kit as the source of truth for SDD workflow. `.specify/` scripts/templates drive the workflow; oMLX only accelerates selected simple outputs. Codex should match the legacy Claude Code routing rules for shared Task Router categories; only the guidance format differs. If `.specify/` is missing, initialize Codex support first: ```bash specify init --here --ai codex --offline ``` For existing Spec Kit projects, prefer repo-local scripts: ```bash .specify/scripts/bash/create-new-feature.sh .specify/scripts/bash/setup-plan.sh .specify/scripts/bash/check-prerequisites.sh .specify/scripts/bash/update-agent-context.sh codex ``` ## Responsibility Split - Codex handles `constitution`, `specify`, `clarify`, `plan`, review, security-sensitive logic, and ambiguous product decisions. - The local model handles `tasks`, `checklist`, `analyze`, summaries, and repetitive transformations through `./scripts/call-omlx.sh`. - Simple implementation tasks such as boilerplate, CRUD, config, docstrings, migration drafts, and test stubs may call oMLX. - Complex implementation tasks such as auth, security, core logic, concurrency, and data-loss risk stay with Codex. - If a preset validator fails, Codex fixes the output directly. ## Commands - Create or update Codex agent context: `.specify/scripts/bash/update-agent-context.sh codex` - Generate with local preset: `./scripts/call-omlx.sh "$(cat plan.md)" --preset speckit-tasks` - Checklist with local preset: `./scripts/call-omlx.sh "$(cat spec.md plan.md)" --preset speckit-checklist` - Analyze with local preset: `./scripts/call-omlx.sh "$(cat spec.md plan.md tasks.md)" --preset speckit-analyze` ## Quality Gates - Specs must be testable, technology-agnostic, and limited to at most three explicit clarification markers. - Plans must identify language, dependencies, storage, project type, and constitution checks. - Tasks must use `T001`, `T002`, and so on; mark parallel work with `[P]` only when files and dependencies are independent. - Implementation must preserve existing user changes and keep patches scoped.

{ "prompt": "# Spec Kit With Codex\n\nUse GitHub Spec Kit as the source of truth for SDD workflow. `.specify/`\nscripts/templates drive the workflow; oMLX only accelerates selected simple\noutputs. Codex should match the legacy Claude Code routing rules for shared Task\nRouter categories; only the guidance format differs.\n\nIf `.specify/` is missing, initialize Codex support first:\n\n```bash\nspecify init --here --ai codex --offline\n```\n\nFor existing Spec Kit projects, prefer repo-local scripts:\n\n```bash\n.specify/scripts/bash/create-new-feature.sh\n.specify/scripts/bash/setup-plan.sh\n.specify/scripts/bash/check-prerequisites.sh\n.specify/scripts/bash/update-agent-context.sh codex\n```\n\n## Responsibility Split\n\n- Codex handles `constitution`, `specify`, `clarify`, `plan`, review,\n security-sensitive logic, and ambiguous product decisions.\n- The local model handles `tasks`, `checklist`, `analyze`, summaries, and\n repetitive transformations through `./scripts/call-omlx.sh`.\n- Simple implementation tasks such as boilerplate, CRUD, config, docstrings,\n migration drafts, and test stubs may call oMLX.\n- Complex implementation tasks such as auth, security, core logic, concurrency,\n and data-loss risk stay with Codex.\n- If a preset validator fails, Codex fixes the output directly.\n\n## Commands\n\n- Create or update Codex agent context:\n `.specify/scripts/bash/update-agent-context.sh codex`\n- Generate with local preset:\n `./scripts/call-omlx.sh \"$(cat plan.md)\" --preset speckit-tasks`\n- Checklist with local preset:\n `./scripts/call-omlx.sh \"$(cat spec.md plan.md)\" --preset speckit-checklist`\n- Analyze with local preset:\n `./scripts/call-omlx.sh \"$(cat spec.md plan.md tasks.md)\" --preset speckit-analyze`\n\n## Quality Gates\n\n- Specs must be testable, technology-agnostic, and limited to at most three\n explicit clarification markers.\n- Plans must identify language, dependencies, storage, project type, and\n constitution checks.\n- Tasks must use `T001`, `T002`, and so on; mark parallel work with `[P]` only\n when files and dependencies are independent.\n- Implementation must preserve existing user changes and keep patches scoped.", "variables": [], "notes": "Use GitHub Spec Kit as the source of truth for SDD workflow. `.specify/` scripts/templates drive the workflow; oMLX only accelerates selected simple outputs. Codex should match the legacy Claude Code routing rules for shared Task Router categories; only the guidance format differs.", "metadata": { "source": { "provider": "github-codex-prompts", "repository": "https://github.com/wewececetw/task-router", "path": ".codex/prompts/speckit.md", "ref": "fe7b5945fd88ae4f28f4437554b690db3aa70613", "url": "https://github.com/wewececetw/task-router/blob/fe7b5945fd88ae4f28f4437554b690db3aa70613/.codex/prompts/speckit.md", "key": "wewececetw/task-router/.codex/prompts/speckit.md" } } }

speckit

About

Spec Kit With Codex

Responsibility Split

Commands

Quality Gates

Install speckit and it becomes one more candidate the router can pick - when it fits.

Use it

Manifest

Reviews