菜单
Package a codebase into a compact, structured context bundle for an AI agent. Use when the user wants to "give the whole repo to an LLM", "pack the codebase", hits context limits sharing a project, or asks for a digest/summary of an unfamiliar repo.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Workflow for building scalable messaging systems on Microsoft Azure using queues, event-driven architecture, pub/sub systems, service buses, and cloud-native communication patterns.
Workflow for implementing observability systems on Microsoft Azure using logging, monitoring, tracing, telemetry, metrics, dashboards, and production-grade cloud diagnostics.
Workflow for creating procedural systems in Blender using Geometry Nodes, procedural materials, node-based modeling, and non-destructive generation techniques.
Workflow for creating precise hard surface models in Blender using clean topology, subdivision workflows, bevel systems, and production-ready modeling techniques.
Workflow for using Claude to generate Blender Python scripts, automate 3D workflows, and create procedural scenes, animations, and assets directly inside Blender.
End-to-end workflow for setting up Blender, rigging a character, and creating a simple animation with a clear output (rendered animation).
| name | repo-context-packaging |
| description | Package a codebase into a compact, structured context bundle for an AI agent. Use when the user wants to "give the whole repo to an LLM", "pack the codebase", hits context limits sharing a project, or asks for a digest/summary of an unfamiliar repo. |
Turn a repository into a single, token-efficient bundle an agent can reason over: a tree, the files that matter, and nothing that wastes the budget. Optimize for signal per token, not completeness.
Skip when the task touches only a few known files — just read those directly.
git ls-files | sed 's|/.*||' | sort -u # top-level layout
git ls-files | wc -l # tracked file count
tokei . 2>/dev/null || cloc . # size by language
Identify the architectural core: entrypoints, primary modules, config that defines structure.
Never include: lockfiles, node_modules/vendor/build output, .git, minified/generated code, binaries, media, snapshots/fixtures, dist/coverage. These are most of the bytes and almost none of the signal.
git ls-files \
| grep -vE '(lock|min\.|\.map$|/dist/|/build/|/vendor/|\.snap$)' \
| grep -vE '\.(png|jpg|jpeg|gif|pdf|zip|woff2?|ttf|mp4)$'
If a tool is available, prefer it: npx repomix or files-to-prompt produce structured bundles with sane defaults — review their exclude list, don't trust it blindly.
State the token budget up front and cut from the bottom tier until it fits.
# <repo> context bundle
## Structure
<tree, depth-limited>
## Architecture (3–6 sentences: how it fits together, entrypoints, data flow)
## Files
### path/to/file
```<lang>
<contents, or a summary line if tier 3>
<globs excluded and why — so the agent knows what it can't see>
The "Omitted" section is mandatory: an agent that knows what it's missing asks for it instead of hallucinating it.
## Quality bar
- Fits the stated budget with the architecture summary intact.
- A reader can locate where any feature lives from tree + summary alone.
- Zero generated/lock/binary content inlined.
- Omissions are listed, not silent.
## Anti-patterns
- Dumping every file until truncation eats the important ones.
- Inlining lockfiles or `dist/` "for completeness".
- Tree only, no architecture prose — structure without meaning.
- Silent omission, so the agent confidently reasons about code it never saw.