trial-verify
When facing >=2 viable options for a non-trivial reversible decision, implement/evaluate ALL of them, verify via independent subagent scoring, and PROCEED with the verified winner without asking the user to choose. Triggered by exception (d) of the 1-question rule.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-10
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- trial-verify
- triggers
- trial-verify, multi-option decision, which option, autonomous option resolution
- changelog
- 2026-09-02 cycle32-p2 — externalize examples/patterns to reference.md (ADR-048)
- description
- When facing >=2 viable options for a non-trivial reversible decision, implement/evaluate ALL of them, verify via independent subagent scoring, and PROCEED with the verified winner without asking the user to choose. Triggered by exception (d) of the 1-question rule.
- token_budget
- 2950
03 SKILL.md 正文
Trial-Verify Protocol
Empirically validated 2026-08-24 (2 trials, subagent verified — detail -> reference.md).
Trigger
=2 viable approaches for reversible decision, blast radius Bajo/Medio (v3 s1), material (>1 file or session impact). Trivial skip.
Process
- ENUMERATE: 2-3 candidates (cap 3).
- PROTOTYPE: concrete artifacts only (text/diff/config), never prose. Over budget? Prototype diff part only.
- VERIFY: independent subagent review — NEVER self-grade alone. Rubric 0-5 each: efficacy | token economy | failure-mode exposure | protocol consistency | learning persistence.
- SELECT: highest verified total wins. Disagreement -> synthesize with BOTH objections as constraints; ledger.
- PROCEED: implement immediately; NO menu. T2+ PEV plan stays single human touchpoint.
- LEDGER: mem_save(topic_key="trial/<topic>", type=decision): What/Options/Scores/Winner/Why/Synthesis.
Hard stops (ask human)
- Irreversible/destructive or blast Alto
- Verification unavailable after 2 attempts -> simplest, confidence:low
- User asks "which do you prefer?" -> recommendation + tradeoffs; agency overrides
Budget caps
Max 3 options | max 2 verifications | abort if >15 tool calls.
Anti-Rationalization
| Rationalization | Red Flag | Verification |
|---|---|---|
| "Skill without verification" | Doing work without checking output format | Output matches skill ## Output contract + file:line citaton |
| "Save time skipping this skill" | Using skill directly without resolving deps | skill-graph resolution + cross-ref check |
| "Output is self-evident" | No file:line or confidence marker | Cite file:line or flag confidence: unvalidated |
Red Flags
- Doing work without checking output format -> STOP, re-read skill
- Second occurrence same rationalization -> force RED zone
Verification
- Output matches skill ## Output contract + file:line citaton
- cross-ref-check.ps1 -> SKILL.md OK
Reference Materials
Externalized to keep <=3KB (ADR-048). Worked examples (trial #1/#2), testing patterns, edge cases/anti-patterns -> docs/skills/trial-verify/reference.md Template: .agents/skills/automejora-analyzer/SKILL.md
Refs
Cross-Refs: triple-verify | testing-strategy
04 同仓库更多 Skills
web-quality-audit
Comprehensive web audit: performance, a11y, SEO, responsive, animation, design tokens.
ui-engine
UI system — Grid/Flexbox/@layer/:has(), container queries, compositor-only animation, OKLCH tokens, component patterns
baseline-ui
Anti-slop UI — layout, typography, responsive, animation, tokens. Use for cleanup or polish.
skill-creator
Create new AI agent skills following Agent Skills spec -- auto-bootstrap unknown tasks, define structure, register in AGENTS.md
senior-engineer
Apply senior engineering competencies — system design, trade-off analysis, delegation, mentoring, and tech debt strategy
prompt-engineering
Design effective prompts using SPEARS framework — scope, principles, examples, assertions, refinements, with ReAct and multi-agent patterns