agentfaultlisted
Install: claude install-skill shuo-young/agentfault
# AgentFault
Build evidence-backed, versioned AI supply-chain graphs and turn scattered GitHub reports into source-linked investigation Findings. The current Codex or Claude session is the reasoning engine; bundled Python validates identities, relationships, citations, scoring, and rendering.
## Modes
| Intent | Mode | Output |
|---|---|---|
| Trace an Agent application across the stack | `map` | Versioned three-tier supply-chain graph |
| Inspect one project and its attached evidence | `dossier` | Components, versions, upstream/downstream, CVEs and reports |
| Trace exposure or downstream adoption | `impact` | Top-down or bottom-up evidence path |
| Find important problems across repositories | `scan` | Ranked failure atlas |
| Contrast two frameworks or repository groups | `compare` | Shared vs unique Findings |
| Re-run a saved scope and show changes | `watch` | New, rising, stable, and cooling findings |
| Turn findings into research/product directions | `opportunities` | Evidence-linked opportunity brief |
Default to `map` for stack, dependency, CVE, or project-landscape requests; default to `scan` for recurring-problem requests. Treat `/agentfault` and plain `agentfault` as aliases.
## Supply-chain workflow
1. Resolve the requested application or project to an exact release, commit,
package, image, model, or protocol identity.
2. Classify Components into S0-S8 by provided capability. A repository may
provide multiple Components in different domains.
3. Extra