| name | entity-linker |
| description | End-to-end entity resolution: reads a text file or pasted content, runs preprocessing, calls resolve_link_text, handles disambiguation via the Disambiguation App (or elicitation fallback), and writes the annotated output. TRIGGERS: "link entities in this", "tag mentions", "normalize names in call", "resolve entities against catalog", "link-file", "link-text".
|
| model | inherit |
| tools | ["Read","Write","Bash"] |
entity-linker Skill
Resolves entity mentions in text inputs (transcripts, emails, docs, Slack) against the entity catalog.
Workflow
- Detect source type — use
detect_source_type() or --type flag override.
- Preprocess — call the appropriate cleaner (ASR, email, markdown, HTML, plain).
- Resolve — call MCP tool
resolve_link_text(text, source_type, options).
- Handle ambiguities — if
ambiguities non-empty and host supports Apps, open Disambiguation App via resolve_disambiguate_app. Otherwise use elicit.disambiguate_span loop.
- Render output — call
resolve_render(resolutions, text, format) (default: markdown).
- Write output — save annotated file to
annotated/ directory next to source.
- Report — show auto-linked count, disambiguation count, new candidates staged.
Options
--format markdown|xml|sidecar — output format (default: markdown)
--type asr|email|markdown|html|plain — force source type
--interactive false — headless mode (queues all ambiguities)
Output Location
<input_dir>/annotated/<filename> for /link-file.
Dependencies
- MCP tools:
resolve_link_text, resolve_disambiguate_app, resolve_render
- Skill:
input-preprocessing (auto-invoked for source type detection)