19 added, 19 removed. Audit A to A.
---
name: person_profile
description: Collate a comprehensive profile on a specific person by searching a given dataset, returning the full synthesized report.
---
## Skill Prompt: `person_profile`
**Objective:** Collate a comprehensive profile on a specific person by searching a given dataset, returning the full synthesized report.
**Inputs:**
* `name`: A string representing the person's name (e.g., `"John Doe"`).
* `dataset_name`: The target dataset to search (e.g., `"fabas"`).
**Procedure:**
1. **Insight File & Caching:**
* Resolve the person to its canonical identifier through the standard person-resolution flow.
* Construct the profile insight with `lib.insights.InsightFile(dataset=dataset_slug, skill="person_profile", model=<model>, identifier=<person_identifier>, subdir=True, config_key=<query_and_instructions>)`.
* Use `insight.find(selection="reusable")` and `insight.content()` to reuse a fresh existing profile when available; otherwise generate the profile and persist it with `insight.save(...)`. Do not hardcode `<REPO_PATH>/insights/...` paths.
2. **Data Retrieval & Synthesis:**
- * If the cache misses, dynamically load the following via `load_repository_config("person_profile")`:
- * **Query:** `config['person_profile']['query']` (The question/prompt string to search for the person).
- * **Instructions:** `config['person_profile']['llm_instructions']` (The strict formatting instructions for the LLM).
- * Invoke `dataset_chat` with the following parameters:
- * `dataset_name=dataset_name`
- * `questions=query.format(name=name)`
- * `llm_instructions=llm_instructions`
- * `return_full_docs=True`
- * `max_chunks=25`
+ * Read the existing manual roster and enrich its people through `LinkedInResolver`.
+ * Synchronize the dataset before checking profile freshness.
+ * Load `query`, `llm_instructions`, and `founder_traits_instructions` via
+ `load_repository_config("person_profile")`; both instruction sections always
+ form the standard prompt and freshness key.
+ * On a cache miss, use `build_person_dossier` for dataset evidence when
+ `include_dataset_context` is enabled, add the resolved LinkedIn payload,
+ and synthesize through `generate_markdown`.
3. **Output Generation:**
* Save every synthesized profile through `InsightFile`.
* `person_profile(...)` returns a flat `list[InsightFile]`.
* `person_profile_as_person_objects(...)` runs the same workflow and returns
the populated `list[Person]` required by person-oriented composition.
**CLI Interface:**
* Expose this skill through the shared slash-command harness.
## Usage
- For composition without public enrichment, call `person_profile(...)` or
- `person_profile_as_person_objects(...)` with `allow_public_sources=False`.
- This requires the existing manual persons roster and skips LinkedIn resolution.
+ Both `person_profile(...)` and `person_profile_as_person_objects(...)` use the
+ same standard generation workflow: read the authoritative manual roster, resolve
+ LinkedIn profiles, gather dataset evidence, and apply the configured biographical
+ and founder-trait instructions. Founder traits are assessed for explicitly
+ identified active founders; unsupported assessments report insufficient information.
A missing roster raises an error directing the caller to run `persons_in_dataset`;
- profile generation never discovers people. Set `assess_founder_traits=True` to append
- the configured narrative N001 assessment for explicitly identified active
- founders. Every profile uses `<identifier>-<model>.md`, with the identifier selected
- by `Person.identifier`: LinkedIn ID, otherwise email address, otherwise full name,
- using the standard slugification. Generation options never extend the filename. Generation
- settings affect cache metadata only; changing them regenerates the same profile
- file. Manual overrides retain their normal precedence.
+ profile generation never discovers people.
+
+ Every profile uses `<identifier>-<model>.md`, with the identifier selected by
+ `Person.identifier`: LinkedIn ID, otherwise email address, otherwise full name,
+ using standard slugification. The retained `include_dataset_context` option affects
+ freshness metadata, never filenames. Manual overrides retain their normal precedence.
```bash
conda run -n sictic-env python -m skills.harness /person_profile "<DATASET_NAME>" "<NAME>"
conda run -n sictic-env python -m skills.person_profile \
--dataset "<DATASET_NAME>" \
--person "<NAME_1>, <NAME_2>"
```
Person discovery is owned by `skills/persons_in_dataset`. This skill uses the library roster reader, then gathers evidence and writes individual profiles. Only an explicit discovery run creates a missing roster.