Skip to content

NetFabric/NetFabric.Numerics / .agents/skills/model-selection

model-selection

Verified

Choose and tune which AI model a custom agent, subagent, or squad node should use. USE FOR: picking a model by task capability tier (general-purpose coding/writing, fast/cheap, deep reasoning/debugging, multimodal); efficiency-first vs. capability-first starting strategies; tuning an effort/reasoning-effort parameter instead of switching models; call cadence vs. provider prompt-cache TTL; cost-tier ranking and cost ceilings; cross-provider or cross-model-family diversity for independent/adversarial review; per-harness model field syntax and resolution order (GitHub Copilot CLI, Claude Code, VS Code custom agents); extending this framework to a new harness. DO NOT USE FOR: authoring the rest of an agent's frontmatter or prompt body (use copilot-cli-custom-agents, claude-code-custom-agents, or the target harness's own skill); designing squad topology (use agent-graph-orchestration); SKILL.md authoring (use create-skill).

FileKindSize
SKILL.mdskill md4.4 KB
references/cache-and-cost.mdmarkdown4.0 KB
references/harness-mapping.mdmarkdown3.2 KB
references/selection-criteria.mdmarkdown4.1 KB

references/harness-mapping.md

View on GitHub
# Harness Field Mapping

The criteria in [selection-criteria.md](selection-criteria.md) and [cache-and-cost.md](cache-and-cost.md) apply everywhere; how you *express* the chosen model differs by harness. Always confirm which harness targets the agent before writing the field — never assume from the current session's own tool.

## Comparison table

| Harness | Model field(s) | Value shape | Resolution order (highest wins) | Fallback-list support |
| --- | --- | --- | --- | --- |
| GitHub Copilot CLI (`.agent.md`) | `model`, `reasoning-effort` | Single string only. Generate the canonical lowercase CLI ID from GitHub's current supported-models table or the CLI picker (for example, `gpt-5.4`, `gpt-5.3-codex`, `claude-sonnet-4.6`). Never generate a bare display label. If compatibility requires display-name syntax, copy the complete picker value exactly, including any provider suffix (for example, `"GPT-5.4 (copilot)"`); this is not the preferred authoring form. | Session model set to `Auto` overrides every agent's own `model:` field | None — an inline array is silently ignored. Approximate resilience via `~/.copilot/settings.json` → `subagents.agents.<name>.model` instead |
| Claude Code (`.claude/agents/*.md`) | `model`, `effort` | Single value only — alias (`sonnet`/`opus`/`haiku`/`fable`), a full model ID, or `inherit` (defaults to `inherit`) | `CLAUDE_CODE_SUBAGENT_MODEL` env var > per-invocation `model` parameter > the subagent's own frontmatter `model` > the main conversation's model | None — single value only |
| VS Code custom agents (`.agent.md`, `target: vscode`) | `model` | Either a single string, or an array tried in order until an available model is found | The array is walked top-to-bottom at invocation time; no session-level override documented | Yes — the one harness among these three with native fallback-list support |

Details and exact syntax always live in that harness's own agent-authoring skill (`copilot-cli-custom-agents`, `claude-code-custom-agents`, or the relevant other harness skill) — treat the table above as a quick comparison, not the authoritative reference for writing the field.

## Extending to a new harness

When authoring for a harness not in the table above, work out these before writing a `model:`-equivalent field:

1. **Single value or list?** Does the field accept only one model, or a prioritized array with automatic fallback?
2. **Separate effort/reasoning field?** Is intelligence-vs-latency tuning a distinct field from the model choice itself, or bundled into model selection only?
3. **Session/parent-level override?** Can a broader setting (a session-wide model, an `Auto`-style mode, an org policy) override or ignore the per-agent field regardless of what's written there?
4. **Native cost/budget field?** Does the harness enforce its own per-agent or per-session cost ceiling, making the cost-ceiling heuristic in [cache-and-cost.md](cache-and-cost.md) unnecessary?
5. **Cross-vendor reach?** Is the field restricted to one vendor's models (like Claude Code), or can it name any provider the harness integrates with?

Record the answers as a new row in the comparison table above and cite that harness's own docs/agent-authoring skill as the source of truth for exact field syntax.
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 322kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 198kUpdated Claude CodeCodexVerifiedInstall