Agent Skills
› mtarcure/claude-vibe-squad
› scope-estimation
scope-estimation
GitHub在分析代码库前测量文件数、大小及最大单文件，确定上下文或调用限制，划分读取层级并声明采样比例与覆盖范围，确保结论基于实际读取量。
Trigger Scenarios
需要评估待分析数据集规模时
制定代码分析计划前
Install
npx skills add mtarcure/claude-vibe-squad --skill scope-estimation -g -y
SKILL.md
Frontmatter
{
"name": "scope-estimation",
"audience": "specialist",
"description": "Use before analyzing any corpus to determine how much can actually be read and how strong later coverage claims may be—measure file count, total bytes, and largest item; name the binding context, tool-call, or time budget; assign full-read, sample, and index-only tiers; and carry numeric denominators into the deliverable."
}
Scope Estimation
Measure the corpus before analyzing it, so the retrieval plan and the confidence claims are sized to what was actually read rather than to what was sampled.
Steps
- Count the corpus before opening it: number of files, total bytes, and the largest single file. An analysis plan written before this count is a guess.
- Convert size to a budget in the units that actually bind — context window, tool calls, and wall-clock — and state which one binds first.
- Classify the corpus into read-fully, sample, and index-only tiers. Record the rule used to assign each tier, not just the assignment.
- Declare the sampling fraction per tier as a number. "Reviewed the codebase" is not a scope statement; "read 34 of 210 files, all 9 entry points" is.
- Identify what the chosen tiers structurally cannot answer, and say so before the analysis rather than after a reader asks.
- Re-measure when the corpus grows mid-task. A scope claim inherited from an earlier, smaller corpus is stale and silently overstates coverage.
- Carry the final numbers into the deliverable so every conclusion inherits an explicit denominator.
Acceptance
- File count, total size, and largest-file size are stated as measured numbers.
- The binding budget is named, and the tier assignment rule is written down.
- Sampling fractions appear as counts with denominators, never as adjectives.
- The deliverable names at least one question the chosen scope cannot answer.
- No coverage claim exceeds the measured read set.
Version History
- d5262e2 Current 2026-09-11 11:25
