emmy-knowledge-cleanup

name	emmy-knowledge-cleanup
description	Use when a human wants to audit or maintain the Emmy Agent Knowledge Store as a whole: find duplicate or overlapping pages and claims, identify stale or contradictory content, recommend labels, and decide whether new child pages should be added before any Confluence cleanup or reorganization.

Emmy Knowledge Cleanup

Use this skill for explicit knowledge-store maintenance. This is a whole-store audit and planning workflow, not ordinary ambient lookup, source ingestion, or queue curation.

Source Of Truth

• Knowledge-store entry point: https://confluenceent.cms.gov/spaces/SFIV/pages/1381557075/Emmy+Agent+Knowledge+Store
• Confluence space: SFIV
• MCP server: cms-atlassian-confluence
• Required user environment variable: CONFLUENCE_PERSONAL_TOKEN

Relationship To Other Skills

• Use emmy-knowledge-store for targeted context lookup, durable capture, and narrow canonical updates.
• Use emmy-artifact-ingest for explicit source ingestion from files, URLs, external Confluence pages, or repository checkouts.
• Use emmy-queue-curate for human review and promotion of existing Knowledge Ingestion Queue entries.
• Use this skill only when the task is about store health, duplication, hierarchy design, label hygiene, stale content, or page organization.

Read references/cleanup-protocol.md before auditing or proposing cleanup.

Cleanup Workflow

1. Read the entry page and inspect its current child pages. Treat the entry page as the only stable root; do not assume the current child list is complete or permanent.
2. Build a lightweight store inventory from page metadata, labels, parent-child relationships, and brief page summaries. Load full page content only for candidate duplicates, contested claims, oversized pages, or proposed cleanup targets.
3. Search for exact and near-duplicate titles, repeated source records, overlapping canonical claims, stale queue entries, contradictory facts, orphan pages, missing labels, and pages whose content has outgrown their current parent.
4. Decide whether a new child page is warranted using the decision rules in references/cleanup-protocol.md.
5. Present a cleanup report in chat with findings, evidence page IDs or URLs, proposed actions, risks, and write requirements.
6. Write nothing to Confluence until the human gives explicit approval for the specific cleanup action.
7. Before any approved write, run the emmy-knowledge-store references/write-safety.md checks and re-read the pages being changed.
8. After any approved write, re-read or diff changed pages and report the final page URLs or IDs.

Cleanup Boundaries

Do not delete Confluence pages. Do not move pages unless a future tool exposes an explicit move operation and the human specifically approves that move. Prefer creating a proposed cleanup plan over performing structural edits.

When cleanup identifies duplicate canonical content, prefer one of these non-destructive actions after approval:

• update the destination page with a concise canonical version;
• add a note pointing duplicate readers to the canonical page;
• add or correct labels;
• create a new child page for a durable topic area;
• queue uncertain consolidation work for human review.

Do not store secrets, local scratch paths, personal data, speculative claims, or uncurated conclusions as cleanup output.

Tool Use

Use Confluence MCP read tools for page, child, label, history, and diff inspection. Use write, comment, label, attachment, or create tools only after explicit approval and only for the approved cleanup action.

Do not use Jira tools from this workflow, even though the shared MCP config may expose them. Do not perform source ingestion from this workflow.