// Entry workflow for LLM engineering tasks (provider integration, calling wrapper, routing profiles, prompt templates, cost/telemetry, credentials/config keys). Routes to one procedure and enforces required verification.
Use only in repos that still include the `init/` bootstrap kit to produce Stage A/B/C artifacts, update skill pack selection, and sync provider skill wrappers.
Create a goal-aligned macro-level roadmap artifact (dev-docs/active/<task>/roadmap.md) by asking clarifying questions and recording assumptions; planning only (no code changes).
Build a complete, production-embedded Agent module for a real feature request (API + optional worker/sdk/cron/pipeline), including blueprint, scaffolded runtime, prompt pack, docs, and registry entry. Enforces explicit user approvals, no-secrets-in-repo, fixed API route names (run/health), and conversation/memory strategy decisions.
Read-only progress reporter. Produces a structured status snapshot from existing project/task artifacts (project hub + dev-docs) and includes an actionable next command; never modifies repo files.
Define and execute a distinctive visual direction for net-new UI or brand refresh work (art direction, typography, color, motion, composition), then deliver a polished reference implementation. Primary intent: creative design direction and visual differentiation; pairs well with style-frontend-ui and build-react-components for engineering hardening.
GitHub Actions CI skill: integrate automated tests (web/api/mobile/perf) with consistent artifacts.
| name | llm-engineering |
| description | Entry workflow for LLM engineering tasks (provider integration, calling wrapper, routing profiles, prompt templates, cost/telemetry, credentials/config keys). Routes to one procedure and enforces required verification. |
.ai/llm-config/registry/* as the SSOT for identifiers and allowed config keys.| Task | Open the procedure |
|---|---|
| Add / integrate a provider | reference/procedures/add-provider.md |
| Standardize or introduce a single calling wrapper | reference/procedures/standardize-calling-wrapper.md |
| Add / change model routing profiles | reference/procedures/add-model-profile.md |
| Add / change prompt templates (versioned) | reference/procedures/update-prompt-template.md |
| Release / major change review | reference/procedures/release-check.md |
Open one example only if you need a concrete template.
| Example | Open |
|---|---|
| Add a provider (OpenAI-style) | reference/examples/e2e-01-add-provider-openai.md |
| Update a prompt template (versioned) | reference/examples/e2e-02-update-prompt-template.md |
| Add a model routing profile | reference/examples/e2e-03-add-model-profile.md |
No secrets in repo
credential_ref).Registry-first configuration
.ai/llm-config/registry/config_keys.yaml.Single calling surface
Versioned prompts
prompt_template_id + immutable version.credential_ref; no secrets)If you changed skills:
node .ai/scripts/lint-skills.mjs --strictnode .ai/scripts/sync-skills.mjs --scope current --providers both --mode reset --yesRegistry sanity (recommended; run before release):
node .ai/skills/workflows/llm/llm-engineering/scripts/validate-llm-registry.mjs--strict in CI.If you changed LLM config/env keys (or introduced new ones):
node .ai/skills/workflows/llm/llm-engineering/scripts/check-llm-config-keys.mjsIf the host repo has tests/lint:
.codex/skills/ or .claude/skills/ directly (generated)..ai/, .system/).