一键导入
atlas-graph
Use when the repository has Atlas graph support enabled and needs a repo knowledge graph generated, refreshed, verified, or reviewed
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。
菜单
Use when the repository has Atlas graph support enabled and needs a repo knowledge graph generated, refreshed, verified, or reviewed
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
Fixture skill with bounded read-only tool access
Malicious fixture skill
Use at session end to capture durable lessons, vocabulary, decisions, and memory promotion proposals in an Atlas workspace
Use when AI context needs cleanup — atlas doctor context-size advisories, an oversized AGENTS.md or CLAUDE.md, bloated memory or vocabulary files, or durable docs that need compaction, relocation, or dedup
Use when a repository needs Atlas setup, repair/update, AGENTS.md refresh, AI memory refresh, vocabulary cleanup, or review after major codebase changes
Use when an AI tool idea, plan, implementation, release, or incident needs an Atlas review — deciding whether to build, planning, challenging a plan or implementation, gating a release or promotion, or running a postmortem
| name | atlas-graph |
| description | Use when the repository has Atlas graph support enabled and needs a repo knowledge graph generated, refreshed, verified, or reviewed |
Use this skill only when graph support is enabled in the Atlas config. The Atlas CLI owns deterministic validation and advisory reporting. This skill owns generator detection, graph generation, metadata writing, and diff review. It never installs generators, never runs network-backed enrichment, and never writes outside config-resolved graph paths.
The initial supported generator is graphify in code-only mode so runs stay offline and deterministic. Treat the graph as a generated artifact: rebuild it, review the diff, and do not hand-edit generated files.
Run the phases below in order.
From the repository root, run Atlas doctor with the local package first:
npx --no-install @blazity-atlas/core doctor
Fall back to the published package only when the local package is not installed:
npx --yes @blazity-atlas/core@latest doctor
Stop before graph work when doctor reports manual conflicts or fixable drift. Structure repair precedes graph generation. If doctor --fix refuses because the worktree is dirty, ask the user whether to commit, stash, or explicitly rerun with --force; do not use --force automatically.
Locate the workspace config: read .ai/config.json, or when absent, follow the .atlas repo-root pointer to <root>/config.json. Resolve every path through that config - never hardcode .ai/.
Continue only when:
features.graph.enabled is true.features.graph.generator.name is graphify.features.graph.generator.version is a non-empty pinned version.If the feature is absent or disabled, stop and explain that graph support is opt-in. Do not edit config unless the user explicitly asks.
Check whether graphify is installed:
command -v graphify
graphify --version
graphify --help
If graphify is missing, refuse politely with install guidance and stop:
Atlas graph support is enabled, but graphify is not installed. Install graphify through your normal Python/toolchain workflow, pin that exact version in features.graph.generator.version, then rerun this skill. I will not install it automatically.
Do not install it automatically.
If the installed version differs from features.graph.generator.version, stop and report the mismatch. Upgrading or downgrading the generator is a config and dependency change that needs review.
Use graphify --help to confirm the installed CLI's code-only/offline and output-directory flags before running it. If the installed CLI cannot run in code-only/offline mode or cannot route every output into the configured graph directory, stop and report the blocker.
Resolve the graph directory from paths.graph; when absent, use the default resolver value graph under artifactRoot. All generated files, including graph.meta.json, must stay under that directory.
Expected sidecar:
{
"generator": {
"name": "graphify",
"version": "<pinned version>"
},
"buildSha": "<git HEAD sha>",
"scope": "code",
"provenance": "extracted"
}
Use scope: "code" for code-only graphify runs. Use provenance: "extracted" for deterministic code facts. Do not use mixed provenance unless the run intentionally includes non-deterministic or model-generated material and the user approved that broader scope.
Before running the generator, capture the current commit:
git rev-parse HEAD
Create the configured graph directory if needed. Run graphify from the repository root with the installed CLI's code-only/offline mode and configured output directory. Do not allow outputs outside the graph directory. Do not run docs/media/LLM enrichment.
After graphify finishes, write or refresh graph.meta.json in the graph directory with the pinned generator version and captured buildSha.
Run atlas doctor again:
npx --no-install @blazity-atlas/core doctor
The graph advisories should be absent for a fresh build. If graph-generator-drift, graph-meta-invalid, or graph-stale remains, report the exact finding and stop.
Show the graph diff for review:
git diff -- <resolved graph directory>
A second run at the same commit should be a no-op. If it is not, report the generated-file instability before suggesting a commit.