AGENT SKILL

constraint-driven-development

Preserve an existing project's quality constraints during implementation or review, especially when checks fail or a change appears to weaken tests, thresholds or safeguards.

License 许可
MIT
所属仓库
b2vv/ai-factory
最近更新
2026-11-15
Stars
★ 0
Forks
0

01 前言 Front Matter

name
constraint-driven-development
description
Preserve an existing project's quality constraints during implementation or review, especially when checks fail or a change appears to weaken tests, thresholds or safeguards.

02 资源文件 3

03 SKILL.md 正文

Constraint-driven development

Read the project's accepted constraints, scripts, task acceptance and applicable instructions first. Reuse their owner and primary task record. This skill is a focused quality-floor check, not another implementation or review pipeline. It does not prescribe a stack, coverage percentage, new CONSTRAINTS.md, hooks, tools or CI.

Before implementation, record the applicable floor and evidence in the existing checkpoint: accepted behavior, relevant checks, permitted exceptions and exact scope. A project with no documented threshold does not inherit an arbitrary threshold from this skill. Ask only when a genuinely unresolved decision affects the authorized work.

On a failing check, investigate the implementation before weakening the check. Inspect changes to thresholds, exclusions, skipped/deleted tests, assertions, suppressions, stubs/empty catches and exception policy. These are signals, not proof: an assertion refactor can preserve or improve behavioral coverage. Compare the old failure-detection capability with the replacement and its relevant execution evidence. A confirmed reduction needs the contract owner's explicit, scoped decision; an author's explanation alone is not that decision.

Bind evidence to repository/root, base, HEAD, and the exact scoped dirty/untracked file hashes. Include deletions and the scope inventory, not only tracked diff. After HEAD or relevant bytes change, recheck the affected evidence; stale review is not inherited automatically. Missing input, skipped checks or unavailable independent review are BLOCKED/NOT_TESTED, never clean.

For interruption or a second sprint task, read the current primary record and establish that task's checkpoint before coding. Verify graph freshness if used; a stale graph can be replaced by direct file/caller inspection with that limitation recorded. Keep the existing reviewer route, local commit gate and final-delta review.

Read floor evidence when assessing a signal or preparing the evidence record, including its Anti-patterns section, and the shared reference anti-patterns (4-field template + cross-cutting rules). The floor moves only by an explicit, reviewed owner decision — never by relaxing a check to obtain PASS. The shared offline validator checks record completeness and snapshot equality; it does not detect all weakening or prove the truth of supplied evidence. Use the project's real tests and independent review for those claims.

Adapted from Addy Osmani's agent-skills at the pinned source recorded in provenance. MIT license.

04 同仓库更多 Skills

b2vv
b2vv/ai-factory

skill-creator

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

★ 0
b2vv
b2vv/ai-factory

skill-creator

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

★ 0
b2vv
b2vv/ai-factory

canvas-design

Create beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.

★ 0
b2vv
b2vv/ai-factory

frontend-design

Visual interface design so it reads as intentional rather than templated. Use when the user builds or reworks the appearance of a web system, portal, app, site or individual screen; asks that it not look AI-generated; or asks about palette, typography, density or design direction. Additional trigger phrases: "avoid an AI-generated look".

★ 0
b2vv
b2vv/ai-factory

ponytail

Lazy senior mode: requires the simplest solution that actually works. First ask whether this work is needed at all (YAGNI), then consider what already exists in the codebase, then stdlib, then a native platform capability, then an already installed dependency — and only then custom code. Three levels: lite, full (default), ultra. Use for any coding task — writing, adding, refactoring, fixing, reviewing, choosing a library or dependency — or on triggers "ponytail", "be lazy", "lazy mode", "yagni", or when the user asks for the simplest solution, less code or no overcomplication, or complains about over-engineering, boilerplate or unnecessary dependencies. Not for non-coding tasks (prose, translation, general knowledge, summaries). Additional trigger phrases: "the simplest solution", "less code", "do not overcomplicate".

★ 0
b2vv
b2vv/ai-factory

plan-critique

Critiques a completed plan through four independent lenses — AI-slop, logical gaps, unnecessary complexity, uncovered edge cases — and consolidates findings into blocking / non-blocking. Use as a gate after writing a plan or spec, before splitting into tasks, or on triggers "critique the plan", "stress-test the plan", "plan critique".

★ 0