critical-reception

name	critical-reception
description	Query Metacritic, GameSpot, and TVGuide (resolved and read via Bright Data) for a personality's existing game/film/TV adaptation, extracting critic/user scores, narrative themes critics highlighted, and reception debates. Use when generate-psych-profile or backlog-ranking want a critical-reception signal on top of gather-citations' facts and fandom-lore's fan interpretation.
metadata	{"agent":"research-factcheck","source":"Project Atlas Agent Skills Manifest (Critical Reception enrichment)","layer":"L5-content-verify"}

critical-reception

Harvest professional critic/audience reception for a personality's existing adaptation and structure it into signal the narrative stage can use -- without ever storing the raw review text.

When to use

• Hidden Legends Discovery flagged an existing_adaptation (title + type) for a discovered personality -- this skill only runs when that field is present. Most discoveries won't have one; skip entirely rather than guessing a source_work.
• Distinct from fandom-lore: this is keyed by the work (the specific film/game/TV series), not the subject directly -- a character doesn't have a Metascore, the release does.

Sources (spec §5.5 mapping, Critical Reception enrichment)

• Bright Data (live) via the brightdata MCP server -- same vendor as gather-citations/ fetch-signals/fandom-lore, no new vendor: search_engine to resolve each of site:metacritic.com, site:gamespot.com, site:tvguide.com for <source_work>, then scrape_as_markdown on whichever resolve (not every work will be covered by all three).

Function signature (manifest contract)

def critical_reception(subject: str, source_work: str) -> dict:
    """Return structured critical/audience reception signal for a personality's adaptation, or
    an empty/zero-confidence result if none of the three sources cover it."""

Inputs / Outputs

• Input: subject, source_work.
• Output: {subject, source_work, sources_resolved[], critic_score?, user_score?, score_divergence_note?, narrative_themes[], reception_debates[], coverage_confidence, provenance: "critical_reception"} -- see hermes/agents/critical-reception/schema.json.

Copyright-safe rule (spec §2.5)

Scraped review/score pages are a transient input to the extraction call only -- never write raw review text to storage. Persist only the structured signal, each theme/debate backed by a short (<25 word) paraphrase in evidence, never a verbatim quote.

Provenance rule

Tagged provenance: "critical_reception" -- a third tier distinct from gather-citations' canon_fact and fandom-lore's fan_interpretation. This is professional critical opinion, not verified fact and not fan headcanon; downstream agents hedge it accordingly ("critics noted...").

Implementation

1. Resolve sources: call brightdata's search_engine three times (site:metacritic.com <source_work>, site:gamespot.com <source_work>, site:tvguide.com <source_work>). Keep whichever of the three actually resolve to that domain -- not all three will exist for every work (e.g. TVGuide won't cover a game).
2. If none resolve, stop here and return {subject, source_work, sources_resolved: [], narrative_themes: [], reception_debates: [], coverage_confidence: 0, provenance: "critical_reception"} directly -- don't invoke the LLM to invent a critic's opinion for a work none of the three cover.
3. Read each resolved page: call brightdata's scrape_as_markdown on each.
4. Call the hermes-bridge MCP tool run_judgment_agent(insight_type="critical-reception", query=<ask>, context={"subject": subject, "source_work": source_work, "sources_resolved": [...], "scraped_content": [...]}). Hits Hermes's POST /v1/agents/critical-reception (hermes/agents/critical-reception/), same self-correcting LLM loop / mock-fallback pattern as gather-citations/fandom-lore -- real retrieved content into context from the start.

Backend dependency

• API keys: Bright Data is live (already wired for gather-citations/fetch-signals/ fandom-lore) -- no new vendor, no new budget line.
• Writing structured signal to the graph (Railway) has the same stubbed status as gather-citations'/fandom-lore's output until the Postgres/AGE backbone is wired.

Model

deepseek-direct/deepseek-chat for extraction. Multiple scraped pages per call -- apply Jina Reader-style compression if combined content exceeds prompt budget (spec §11.3).