Stats
Actions
Tags
From te
Use this skill when picking a model for a subagent or task — Haiku for mechanical work (search, test runners, log triage, scaffolds), Sonnet for judgment (debugging, architecture, multi-file changes), Opus only when Sonnet stalls. Apply at the start of any session, before spawning a subagent, or whenever a task's complexity is unclear.
How this skill is triggered — by the user, by Claude, or both
Slash command
/te:tier-modelThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Most work in a session does not require the largest model. The supervisor often *is* Sonnet or Opus — fine, that's where judgment lives — but the subagents it spawns should usually be Haiku. Running every subagent on the supervisor's tier is the silent budget killer.
Most work in a session does not require the largest model. The supervisor often is Sonnet or Opus — fine, that's where judgment lives — but the subagents it spawns should usually be Haiku. Running every subagent on the supervisor's tier is the silent budget killer.
| Task | Model |
|---|---|
| File search, grep, path discovery | Haiku |
| Reading a file and summarizing it | Haiku |
| Running and interpreting tests/builds/linters | Haiku |
| Log triage, error extraction from CI | Haiku |
| Mechanical refactor with an explicit spec | Haiku |
| Single-file edits under a clear contract | Haiku |
| Scaffolds following an existing pattern | Haiku |
| Doc/API lookup and snippet extraction | Haiku |
| Verifier subagents (did the checks pass?) | Haiku |
| Debugging an unknown root cause | Sonnet |
| Architecture or design decisions | Sonnet |
| Multi-file changes requiring consistency | Sonnet |
| Code review with judgment (taste, trade-offs) | Sonnet |
| Performance investigation | Sonnet |
| Cross-system reasoning where Sonnet stalled twice | Opus |
| Novel algorithms or genuinely hard correctness work | Opus |
If unsure, start with Haiku. Escalate only if it returns STATUS: needs_input or visibly flounders.
Smaller models on focused tasks tend to return tighter outputs. A Haiku subagent told to "find where session expiry is enforced and return file:line" returns three lines. The same prompt to Opus often returns three lines plus unsolicited analysis and suggestions — which then enter the supervisor's window. Right-sizing the model is right-sizing the reply.
Provides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.
npx claudepluginhub muktadirhassan/skills --plugin te