Documentation
¶
Overview ¶
Package skills loads and serves skill guides — Markdown documents that instruct LLM agents how to compose chip's tools for common Collibra workflows. Skills are embedded in the binary at build time and exposed via two MCP tools: list_collibra_skills and load_collibra_skill.
Index ¶
- Constants
- func Enabled(toolConfig *chip.ServerToolConfig) bool
- func NewListTool(catalog *Catalog) *chip.Tool[listInput, listOutput]
- func NewLoadTool(catalog *Catalog) *chip.Tool[loadInput, loadOutput]
- func RegisterAll(server *chip.Server, externalDir string) error
- type Catalog
- type Resource
- type Skill
Constants ¶
const FeatureName = "skills"
FeatureName is the experimental-feature identifier used to gate skills.
const Instructions = `# Collibra MCP Server
This server connects you to Collibra, an enterprise data governance platform — the authoritative catalog for what data the organization has, what it means, how it relates, and who governs it. Reach for these tools when the user asks about **discovering, understanding, or governing data**: "what customer data do we have?", "what does this metric measure?", "which columns hold PII?", "where does this KPI come from?". These tools operate on metadata and governance, not on the underlying row-level data.
## Key concepts
- **Assets**: any item in the catalog — its ` + "`assetType`" + ` (Table, Column, Business Term, KPI, Report, Policy, …) determines what it represents.
- **Business Terms**: standardized definitions for business concepts.
- **Data Contracts**: agreements defining data product interfaces.
- **Classifications**: data classes applied to assets to categorize them — most commonly for data sensitivity (PII, PHI), but also for arbitrary taxonomies.
## Skills
This server ships skill guides — short Markdown docs that document the right tool sequences, ID-bridging rules (e.g. lineage entity IDs are not DGC asset UUIDs), permission requirements, and known limitations (e.g. column-level lineage) for each Collibra domain (discovery, lineage, asset create/edit, data products, classification). Skill content is not visible in tool names or schemas.
Before composing a multi-step workflow, run skill discovery: call ` + "`list_collibra_skills`" + ` (optionally with a ` + "`query`" + ` matching the user's intent) and load the matching ` + "`collibra/*`" + ` skill via ` + "`load_collibra_skill`" + `. Start with ` + "`collibra/index`" + ` if unsure which skill applies. Load related skills the index or a loaded skill points to when they apply.
Exceptions: a single tool call whose intent is obvious from the tool's own description does not need a skill. A follow-up call in a domain whose skill you have already loaded this session does not need another discovery round. If no skill clearly matches, proceed with the tool descriptions.
When a tool returns a permission error, surface to the user which scope is needed (e.g. ` + "`dgc.ai-copilot`" + `, ` + "`dgc.classify`" + ` + ` + "`dgc.catalog`" + `).`
Instructions is the full initialize-instructions text shipped when the skills feature is enabled. It replaces the default navigator (pkg/chip/instructions.md) instead of appending to it: with skills on, the workflow recipes, tool categories, and key patterns the default navigator carries also live inside the skill bodies, so eagerly shipping them duplicates context and competes with the "discover skills first" directive. The text mirrors Datadog's pattern — orientation, the existence-of-skills claim, the discovery protocol, and the fall-back exception — and deliberately omits per-tool recipes.
Variables ¶
This section is empty.
Functions ¶
func Enabled ¶
func Enabled(toolConfig *chip.ServerToolConfig) bool
Enabled reports whether the skills feature is active for the given config. Both tool names must be enabled; the bootstrap snippet describes a workflow that needs both list_collibra_skills and load_collibra_skill, so disabling either takes the feature down as a unit. This is the canonical gate — callers use it for both the instructions snippet and tool registration so they cannot diverge.
func NewListTool ¶
NewListTool returns the list_collibra_skills tool wired to the given catalog.
func NewLoadTool ¶
NewLoadTool returns the load_collibra_skill tool wired to the given catalog.
func RegisterAll ¶
RegisterAll loads the embedded catalog, optionally overlays it with skills from externalDir, and registers the two skill tools. An empty externalDir means embedded-only. Callers must check Enabled first; RegisterAll does not re-gate.
Types ¶
type Catalog ¶
type Catalog struct {
// contains filtered or unexported fields
}
Catalog is the in-memory index of all skills. Built once at startup.
func LoadWith ¶
LoadWith returns the embedded catalog, optionally overlaid with skills loaded from externalDir. Skills in externalDir whose name (e.g. "collibra/lineage") matches an embedded skill replace it wholesale — body, description, related, and resources all come from the external entry. New names are added. An empty externalDir is equivalent to Load.
type Resource ¶
Resource is a bundled file (e.g. references/column-lineage-workaround.md) that a skill can offer for progressive disclosure.
type Skill ¶
type Skill struct {
Name string
Description string
Related []string
Body string
Resources []Resource
}
Skill is one entry in the catalog. A skill bundles a Markdown body, a short description, optional cross-references to related skills, and an optional set of reference files that can be loaded on demand.
func (*Skill) Resource ¶
Resource looks up a bundled resource on a skill by its relative path (e.g. "references/column-lineage-workaround.md").
func (*Skill) ResourcePaths ¶
ResourcePaths returns the list of relative paths for bundled resources.
Source Files