スキル一覧に戻る
my-entourage

grounded-query

by my-entourage

MyEntourage.dev agent skills to have all of your AI agent context across communication channels in one place in sync.

2🍴 0📅 2026年1月21日
GitHubで見るManusで実行

SKILL.md


name: grounded-query description: Verifies factual claims against source documents. Use when answering questions about meeting content, decisions, project details, or any assertion that should be grounded in data files.

Purpose

Ensure all factual claims in responses are supported by evidence from source documents. Prevents hallucination and unsupported status claims.

When to Use

Automatically apply this workflow when:

  • Answering questions about what was discussed, decided, or planned
  • Making claims about project status, timelines, or deliverables
  • Summarizing meeting content or extracting information from transcripts
  • Any query where accuracy matters and source files exist

Workflow

Step 1: Generate Initial Response

Answer the user's query based on your understanding of the data files.

Step 2: Extract Verifiable Claims

Identify statements in your response that are factual assertions, such as:

  • "The team decided to use X"
  • "Feature Y is complete"
  • "Person Z mentioned..."
  • Dates, names, decisions, status claims

Step 3: Search for Evidence

For each claim, search the data directory for supporting evidence:

  • Use Grep to find relevant mentions
  • Use Read to examine context around matches
  • Note the file path and relevant line/section

Step 4: Classify Evidence

For each claim, assign a status:

  • SUPPORTED - Direct evidence found in source files
  • PARTIAL - Related mentions but not exact confirmation
  • NOT FOUND - No evidence located (doesn't mean false, just unverified)

Step 5: Output Format

Always end your response with an evidence table:

### Evidence

| Claim | Status | Source |
|-------|--------|--------|
| [Brief claim] | SUPPORTED/PARTIAL/NOT FOUND | folder/filename |

If evidence is NOT FOUND for key claims, add a note:

Note: Some claims could not be verified against available data files. Confidence is lower for unverified items.

Example

Query: "What database did the team choose?"

Response: The team chose Supabase as the primary database, with pgvector for embeddings support.

Evidence

ClaimStatusSource
Supabase as primary databaseSUPPORTEDgranola/2025-12-22T02_50_00Z.md
pgvector for embeddingsSUPPORTEDapple-note-voice-memos/2025-12-21T04_48_00Z.md

Important

  • Do NOT claim something is "Done" or "In Progress" based only on meeting discussions
  • Distinguish between "discussed/triage" and "implemented"
  • When in doubt, mark as PARTIAL or NOT FOUND rather than SUPPORTED

After Output

This skill returns results to the calling context. Do not stop execution. Continue with the next step in the workflow or TODO list.

スコア

総合スコア

60/100

リポジトリの品質指標に基づく評価

SKILL.md

SKILL.mdファイルが含まれている

+20
LICENSE

ライセンスが設定されている

0/10
説明文

100文字以上の説明がある

+10
人気

GitHub Stars 100以上

0/15
最近の活動

3ヶ月以内に更新がある

0/10
フォーク

10回以上フォークされている

0/5
Issue管理

オープンIssueが50未満

+5
言語

プログラミング言語が設定されている

+5
タグ

1つ以上のタグが設定されている

0/5

レビュー

💬

レビュー機能は近日公開予定です