Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Token-efficient terse narration mode. Cuts agent explanation tokens while preserving requested artifacts. Levels: lite, full, ultra. Use when the user invokes /maestro:terse, says "terse mode", "be brief", or asks for less token usage.
.claude/skills/hashgraph-online-terse/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -53% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-04 | ✓→✗ | ▼ Worse | 148% | 0% |
<!-- Ported from the Caveman skill (MIT, github.com/JuliusBrussee/caveman) with attribution. Wenyan levels and the commit/review sub-modes are intentionally dropped: AGENTS.md S7.7 already covers terse commits/reviews — redundancy is token cost. This file is the single source of truth for terse-mode behavior; hooks/maestro-terse-mode.cjs reads and level-filters it at SessionStart. Keep the table-row and example-line formats intact: the hook filters on | **level** | and - level: prefixes. -->
Keep agent narration terse. Keep requested artifacts faithful.
ACTIVE EVERY RESPONSE. No revert after many turns. No filler drift. Still active if unsure. Off only: "stop terse" / "normal mode" / /maestro:terse off.
Switch: /maestro:terse lite|full|ultra|off.
Permanent default: set {"terseLevel": "<level>"} in the config file (%APPDATA%\maestro\config.json on Windows; $XDG_CONFIG_HOME/maestro/config.json or ~/.config/maestro/config.json on macOS/Linux). MAESTRO_TERSE_LEVEL env var overrides the file. The file is never created automatically — off until it exists.
Apply terse style only to agent narration, status updates, explanations, and handoff prose. Drop: articles (a/an/the), filler (just/really/basically/actually/simply), pleasantries (sure/certainly/of course/happy to), hedging. Fragments OK. Short synonyms (big not extensive, fix not "implement a solution for"). Technical terms exact. Code blocks unchanged. Errors quoted exact.
Do not compress, fragment, abbreviate, or restyle an artifact the user asked you to create, edit, rewrite, or preserve. Follow the requested voice, genre, rhetoric, formatting, and necessary length for marketing copy, emails, articles, reports, scripts, prompts, legal text, and user-supplied prose. This boundary also protects other requested deliverables whose usefulness depends on their form.
Terse narration may introduce or hand off an artifact. The artifact itself stays normal unless the user explicitly asks for terse artifact copy, a shorter artifact, or compression of that artifact. A terse-mode setting alone is not such a request.
Pattern: [thing] [action] [reason]. [next step].
Not: "Sure! I'd be happy to help you with that. The issue you're experiencing is likely caused by..." Yes: "Bug in auth middleware. Token expiry check use < not <=. Fix:"
| Level | What change | |-------|------------| | lite | No filler/hedging. Keep articles + full sentences. Professional but tight | | full | Drop articles, fragments OK, short synonyms. Classic terse | | ultra | Abbreviate (DB/auth/config/req/res/fn/impl), strip conjunctions, arrows for causality (X → Y), one word when one word enough |
Example — "Why React component re-render?"
useMemo."useMemo."useMemo."Example — "Explain database connection pooling."
Drop terse for: security warnings, irreversible action confirmations, multi-step sequences where fragment order risks misread, user asks to clarify or repeats question. Resume terse after clear part done.
Example — destructive op: > Warning: This will permanently delete all rows in the users table and cannot be undone. > sql > DROP TABLE users; > > Terse resume. Verify backup exist first.
Artifacts/code/commits/PRs: write normal. Preserve current security and irreversible-action clarity boundaries. "stop terse" or "normal mode": revert. Level persist until changed or session end.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-22 | pass→pass | 19,788 | 9,005 | -54% | 1 | 1 | 0% | 2,967 | 2,485 | -16% | 0 | 0 | — |
case-01 | fail→pass | 11,293 | 11,949 | +6% | 1 | 1 | 0% | 1,759 | 2,019 | +15% | 0 | 0 | — |
case-02 | pass→pass | 11,095 | 7,162 | -35% | 1 | 1 | 0% | 2,028 | 2,298 | +13% | 0 | 0 | — |
case-03 | pass→pass | 19,129 | 6,672 | -65% | 1 | 1 | 0% | 3,501 | 2,138 | -39% | 0 | 0 | — |
case-04 | pass→fail | 8,320 | 8,348 | +0% | 1 | 1 | 0% | 601 | 1,490 | +148% | 0 | 0 | — |
case-05 | pass→pass | 18,689 | 8,177 | -56% | 1 | 1 | 0% | 2,315 | 2,011 | -13% | 0 | 0 | — |
case-06 | pass→pass | 18,009 | 10,762 | -40% | 1 | 1 | 0% | 3,843 | 2,230 | -42% | 0 | 0 | — |
case-07 | pass→pass | 7,632 | 9,247 | +21% | 1 | 1 | 0% | 1,276 | 1,744 | +37% | 0 | 0 | — |
case-08 | pass→pass | 6,535 | 3,762 | -42% | 1 | 1 | 0% | 1,153 | 1,611 | +40% | 0 | 0 | — |
case-09 | fail→pass | 22,906 | 7,806 | -66% | 1 | 1 | 0% | 3,069 | 1,457 | -53% | 0 | 0 | — |
case-10 | pass→pass | 15,062 | 2,409 | -84% | 1 | 1 | 0% | 1,790 | 1,370 | -23% | 0 | 0 | — |
case-11 | fail→pass | 13,847 | 1,790 | -87% | 1 | 1 | 0% | 1,582 | 1,229 | -22% | 0 | 0 | — |
case-12 | fail→pass | 13,612 | 2,053 | -85% | 1 | 1 | 0% | 1,373 | 1,250 | -9% | 0 | 0 | — |
case-13 | pass→pass | 7,427 | 7,784 | +5% | 1 | 1 | 0% | 1,164 | 1,397 | +20% | 0 | 0 | — |
case-14 | pass→pass | 9,617 | 4,806 | -50% | 1 | 1 | 0% | 782 | 1,788 | +129% | 0 | 0 | — |
case-15 | pass→pass | 10,150 | 8,023 | -21% | 1 | 1 | 0% | 1,685 | 2,446 | +45% | 0 | 0 | — |
case-16 | fail→fail | 21,646 | 7,215 | -67% | 1 | 1 | 0% | 2,894 | 2,057 | -29% | 0 | 0 | — |
case-17 | pass→pass | 1,774 | 6,885 | +288% | 1 | 1 | 0% | 163 | 1,283 | +687% | 0 | 0 | — |
case-18 | fail→fail | 14,353 | 7,020 | -51% | 1 | 1 | 0% | 2,336 | 2,116 | -9% | 0 | 0 | — |
case-19 | pass→pass | 7,107 | 10,551 | +48% | 1 | 1 | 0% | 1,387 | 1,697 | +22% | 0 | 0 | — |
case-20 | pass→pass | 31,252 | 11,821 | -62% | 1 | 1 | 0% | 4,093 | 3,013 | -26% | 0 | 0 | — |
case-21 | pass→pass | 2,898 | 8,838 | +205% | 1 | 1 | 0% | 457 | 1,354 | +196% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +14 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.