skill-049

Enterprise Artifact Search Skill (Robust)

This skill delegates multi-hop artifact retrieval + structured entity extraction to a lightweight subagent, keeping the main agent’s context lean.

It is designed for datasets where a workspace contains many interlinked artifacts (documents, chat logs, meeting transcripts, PRs, URLs) plus reference metadata (employee/customer directories).

This version adds two critical upgrades:

1. Product grounding & anti-distractor filtering (prevents mixing CoFoAIX/other products when asked about CoachForce).
2. Key reviewer extraction rules (prevents “meeting participants == reviewers” mistake; prefers explicit reviewers, then evidence-based contributors).

---

When to Invoke This Skill

Invoke when ANY of the following is true:

1. The question requires multi-hop evidence gathering (artifact → references → other artifacts).
2. The answer must be retrieved from artifacts (IDs/names/dates/roles), not inferred.
3. Evidence is scattered across multiple artifact types (docs + slack + meetings + PRs + URLs).
4. You need precise pointers (doc_id/message_id/meeting_id/pr_id) to justify outputs.
5. You must keep context lean and avoid loading large files into context.

---

Why Use This Skill?

Without this skill: you manually grep many files, risk missing cross-links, and often accept the first “looks right” report (common failure: wrong product).

With this skill: a subagent:

• locates candidate artifacts fast
• follows references across channels/meetings/docs/PRs
• extracts structured entities (employee IDs, doc IDs)
• verifies product scope to reject distractors
• returns a compact evidence map with artifact pointers

Typical context savings: 70–95%.

---

Invocation

Use this format:

Task(subagent_type="enterprise-artifact-search", prompt="""
Dataset root: /root/DATA
Question: <paste the question verbatim>

Output requirements:
- Return JSON-ready extracted entities (employee IDs, doc IDs, etc.).
- Provide evidence pointers: artifact_id(s) + short supporting snippets.

Constraints:
- Avoid oracle/label fields (ground_truth, gold answers).
- Prefer primary artifacts (docs/chat/meetings/PRs/URLs) over metadata-only shortcuts.
- MUST enforce product grounding: only accept artifacts proven to be about the target product.
""")