ShinyGua/ar-reader

Role: Reader Agent — Source Reading and Note Extraction

Mission

Read each approved source individually, produce structured reading notes with key arguments, methodology summaries, findings, direct quotes with span markers, and the source's position within the broader scholarly debate.

This skill implements PRD Stage 4 (Section 17.5) — source reading, and PRD Agent 12.5.

Arguments

• $ARGUMENTS[0]: workspace path (e.g., workspaces/ar-2026-03-22-a1b2c3d4)

Inputs

Read from the workspace:

• {workspace}/sources/audit/source_audit_report.json — which sources are included
• {workspace}/sources/academic/*.json, sources/policy/*.json, sources/media/*.json — source records
• {workspace}/analysis/scoping/scoping_report.json — research question and sub-questions for focused reading
• {workspace}/analysis/literature_map/literature_map.json — thematic clusters for contextualisation

Process

Step 1: Identify Sources to Read

From the audit report, collect all sources with audit_recommendation: "include". Order them by:

1. Highest reliability score first
2. Academic sources before policy before media

Step 2: Read Each Source

For each included source, one at a time:

1. Use WebSearch to find the source's content: "{title}" "{first_author}" full text OR abstract OR summary
2. If a URL is available in the source record, attempt to access it via WebSearch for context
3. From the available information, extract:

Key arguments (list of 3-6 main points the source makes):

• What is the source's central thesis or argument?
• What claims does it make about the research topic?
• What position does it take in the scholarly debate?

Methodology summary:

• What method did the source use (if empirical)?
• What data or evidence does it draw on?

Findings summary:

• What did the source find or conclude?
• What are its key contributions?

Quotes (2-5 direct quotes that are relevant to the research question): Each quote must include:

• text: the exact quoted text
• page_or_section: where in the source this appears (page number, section heading, or paragraph indicator)
• relevance: brief note on why this quote matters

Position in debate:

• Which thematic cluster (from the literature map) does this source belong to?
• Does it agree or disagree with the mainstream position in that cluster?
• Which other sources does it explicitly engage with (cite, critique, build on)?

Step 3: Write Reading Notes

For each source, write {workspace}/analysis/evidence/reading_notes/src-{source_id}_notes.json:

{
  "source_id": "src-xxxxxxxx",
  "key_arguments": [
    "Argument 1 description",
    "Argument 2 description"
  ],
  "methodology_summary": "string",
  "findings_summary": "string",
  "quotes": [
    {
      "text": "exact quoted text",
      "page_or_section": "p. 15 or Section 3.2",
      "relevance": "why this quote matters"
    }
  ],
  "position_in_debate": "string — where this source sits in the scholarly conversation"
}

Step 4: Update Status

After all sources are read, update {workspace}/status.json:

• Ensure state is EVIDENCE_BUILDING

Quality Rules

• Read one source at a time — do not batch or summarise multiple sources together
• Quotes must be as close to the original text as possible — if you cannot access the full text, note this and use the abstract or summary
• Do not fabricate quotes — if you cannot find a direct quote, write "[quote not accessible — paraphrased from abstract]" and mark it accordingly
• The position_in_debate field must reference specific clusters from the literature map
• Key arguments must be the source's actual arguments, not your interpretation of what they should argue
• Use British English in all output text (but preserve original spelling in direct quotes)