Asset Graph Management

Asset Graph Management

Use this skill to turn scattered infrastructure and operational knowledge into a reusable asset graph. The graph should be safe to share, easy for agents to update, and grounded in evidence rather than guesses.

Core Workflow

1. Treat any user-provided operational entrypoint or configuration as a discovery seed, not only SSH.
2. Classify the seed and the asset being discussed, then choose the closest generic type.
3. Create or update one record under resources///index.md.
4. Put durable structured facts in frontmatter and reviewable operating notes in Markdown.
5. Represent relationships with stable asset IDs, not display names, URLs, or ports.
6. Separate confirmed facts, clues, gaps, and operations; do not promote guesses into the graph.
7. Store reusable read-only commands in the asset's local scripts/ directory when needed.
8. Keep sensitive or organization-specific values out of public records, examples, logs, and scripts.
9. Rebuild the catalog and run validators before publishing or handing off the package.

What To Read

• For the asset model and record format, read references/asset-model.md.
• For DevOps discovery steps, read references/devops-discovery.md.
• For safety, evidence, and secret-handling rules, read references/evidence-and-safety.md.
• For script entrypoint rules, read references/script-requirements.md.
• Before publishing or sharing the package, read references/publishing-checklist.md.
• To see a complete sanitized example, inspect demo/resources/ and open ui/index.html.

Required Record Shape

Every asset record is a Markdown file with YAML frontmatter. Keep the shape small and stable:

---
asset:
  id: acme-checkout-api
  type: repository
  name: Acme Checkout API
  environment: shared
  status: active
  owners:
    - platform-team
  last_verified:
    at: 2026-01-01
    by: demo-agent
    evidence:
      - "Generated fictional public demo record."
  relationships:
    runs_on:
      - acme-prod-api-01
  evidence:
    - "Demo source: public example generated for this skill."
---

# Acme Checkout API

Short summary, operating notes, and links to scripts or references.

Use unknown for unknown facts. Do not invent owners, dependency edges, endpoints, regions, or status values.

Operating Boundary

Read-only discovery, file inspection, validation, and catalog building are allowed by default and should proceed proactively after the user provides an asset, entrypoint, or access configuration. Entrypoints include SSH targets, URLs, API base URLs, MCP server configs, Jenkins or other CI job URLs, Git remotes, database connection details, object storage endpoints, cloud consoles, SaaS admin URLs, and monitoring dashboards. Restarting services, triggering builds, calling mutating API or MCP tools, deploying code, mutating data, changing DNS/firewalls/IAM, rotating credentials, sensitive reads, or touching live production state requires explicit user authorization and an operation log.

For open-source packages, publish only fictional demo assets, reserved example domains, placeholder repository URLs, and sanitized scripts. If a private organization uses this methodology internally or during evaluation, put real-target artifacts outside the public skill repository, label catalogs as private/eval output, and keep real endpoints and credentials out of the public package.

Demo And Tools

This skill includes a self-contained demo graph with no private assets:

python scripts/validate_demo.py demo
python scripts/build_catalog.py demo demo/catalog/catalog.json --generated-from "fictional public demo records"
python scripts/check_public_safety.py .
python scripts/validate_skill_package.py .

Open ui/index.html in a browser to inspect the demo graph. The UI reads the bundled demo JSON by default, accepts ?catalog=, and can import a local catalog JSON file.

Publication Boundary

This package is designed for open publication. Put methodology in references/, deterministic checks in scripts/, fictional examples in demo/, and browser presentation in ui/. Do not include real hosts, IP addresses, repository URLs, customer names, internal system names, cloud account IDs, local machine paths, tokens, or credentials.