Extract product mentions, score recommendations, and track trends from podcast transcripts. Built for the agent-to-agent economy.
Four tools with complete input schemas and detailed descriptions (avg 180 chars), but significant gaps in output documentation and error handling. All tools accept optional API keys as parameters (security anti-pattern). Naming is clear and verb-driven. Parameter descriptions are thorough with constraints. However, no documented output schemas, no error recovery guidance, and no tool annotations. Descriptions are LLM-optimized but lack dependency hints and when-to-use guidance.
Identify and score podcast sponsor segments (host-read ads, mid-roll, pre-roll). Returns sponsor name, ad placement type, call-to-action URL, estimated read-through rate, and revenue metrics based on category benchmarks — not live ad-platform data. Use for advertising intelligence, CPM estimation, and sponsor outreach research. Use this tool when you need sponsor metrics only; for the full product and recommendation list use extract_podcast_products instead. Reuses cached extraction when episode_id matches a prior extract_podcast_products call. Example: episode_id='huberman-ep-301' → returns [{sponsor:'Athletic Greens', placement:'pre-roll', estimated_cpm:25, read_through_rate:0.72}].
Cross-show product ranking with entity resolution — identify which products are mentioned across multiple podcasts, rank by frequency, and resolve entity aliases (e.g. 'AG1' = 'Athletic Greens'). Returns product name, show count, total mentions, and category. Use for multi-show sponsorship packages, product market penetration analysis, and influencer network mapping. Requires prior extract_podcast_products calls across multiple shows. Example: shows=['huberman-lab','tim-ferriss','lex-fridman'] → returns [{product:'Whoop', show_count:3, total_mentions:12, category:'physical_goods'}].
Extract affiliate products, sponsored brands, and host recommendations from a podcast transcript (text input only — does not process audio or video). Returns product name, category, confidence score, recommendation strength, and sponsor flag. Call this before analyze_episode_sponsors or track_product_trends — both tools reuse its cache. Use for podcast monetization, affiliate program discovery, and episode-to-episode trend tracking. Returns empty products array (not an error) when no product mentions are found. Example: episode_id='huberman-ep-301', category_filter=['physical_goods','supplement'].
API keys exposed as tool parameters (api_key in all 4 tools). Credentials must use server-side injection via environment variables, not tool parameters. Agent traces log all parameters, secrets leak into logs and prompt history.
No documented output schemas. Tool descriptions mention return fields (product name, category, confidence score, sponsor flag, etc.) but no formal schema definition. LLMs cannot plan downstream calls or extract structured data without knowing response shape.
No error recovery guidance. Error responses (e.g., 'Free tier exhausted', 'Invalid API key') do not tell the LLM what to do next. Missing actionable error messages like 'Try search_users() first' or 'Retry with exponential backoff'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Compare product mentions across multiple episodes to identify rising/falling trends, seasonal patterns, and category shifts. Returns trend direction (rising/stable/falling), velocity (mentions per episode), and confidence. Reuses cached extractions from prior extract_podcast_products calls — pass episode_ids that were previously extracted. Use for competitive intelligence, affiliate program ROI tracking, and sponsor negotiation. Example: episode_ids=['ep-301','ep-302','ep-303'] → returns [{product:'Whoop', trend:'rising', velocity:0.33, confidence:0.85}].
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All tools are read-only but this is not declared in the schema. Agents cannot distinguish safe-to-retry tools from destructive ones without annotations.
Missing dependency hints in descriptions. Tools mention caching and reuse (e.g., 'Reuses cached extraction when episode_id matches') but do not explicitly state 'Call extract_podcast_products first' or 'Requires prior extract_podcast_products calls'. Agents may call tools in wrong order.