med-event-takeaways
Post-event analyst note (conferences, KOL events, site visits) with the professional note anatomy — headline verdict, per-stock takeaways with model numbers, KOL distillation, cross-trial comparison lattices, modeled deltas vs consensus (peak × POS), three what's-changed vectors,
Install
npx skills add https://github.com/agentii-ai/agentii-investment-intelligence/tree/main/plugins/vertical-plugins/bio-pharm/skills/agentii/med-event-takeaways
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install agentii-ai-agentii-investment-intelligence@llmmart
git clone https://github.com/agentii-ai/agentii-investment-intelligence.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole agentii-ai/agentii-investment-intelligence collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Methodology inspired by publicly taught sell-side note-writing practice; all text is an original paraphrase.
Defaults
| Parameter | Default Value | Rationale |
|---|---|---|
| note_anatomy | 9 sections | headline → per-stock → KOL → lattices → deltas → vectors → risks → trackers |
| delta_framing | peak × POS | Modeled deltas vs consensus are unadjusted peak × POS |
| conflict_policy | materiality-rated | Surface verbatim, rate materiality, or defer to an outcomes trial |
| vector_set | estimates/thesis/positioning | The three what's-changed vectors per covered name |
Preflight
Run canonical pre-flight per contracts/preflight.md. Propagate X-Agentii-Trace per contracts/x-agentii-trace-header.md. Confirm ticker resolution via search_companies before event queries.
Triggers
- "Write the note on [conference / KOL event]."
- "What are the takeaways from [event] for my coverage?"
- "Distill the [conference] KOL panel on [drug/class]."
- "What changed for [ticker] after [event]?"
- "Model the delta to consensus from [event]'s data."
- "Which stocks move on [event] and why?"
- "Conflict check: what did KOLs disagree on at [event]?"
- "What do I update first — estimates, thesis, or positioning?"
- "Summarize the cross-trial comparisons from [event]."
- "Upside and downside risks after [event]."
- "Link the living trackers that [event] should update."
Production Grounding
- Note anatomy is fixed: headline verdict (paragraph one) → per-stock takeaways with model numbers → KOL distillation → raw data + cross-trial lattices → modeled deltas vs consensus (unadjusted peak × POS) → three what's-changed vectors (estimates / thesis / positioning) → risk bullets → living-tracker links.
- Conflict rule: conflicting views surfaced verbatim, rated by materiality ("incremental positive, not narrative-changing") or deferred to a larger outcomes trial — never silently averaged.
- Consensus delta: model-vs-consensus always explicit; POS changes logged with reasons; abstain where unquantifiable.
- Buy-side lens: client-question section, where investor attention sits, per-name bull/bear pivot conditions.
- Badges:
[FACT]/[DEDUCTED]/[VIEW]with the summary table.
Data Source Priority
- Event content: transcripts/slides via
search_documents+list_sources+read_source_outline/read_source_pages. - Trial context:
search_clinical_trials/get_clinical_trialfor the studies discussed. - Consensus:
search_earnings_calendar(company-level); product consensus derived and labeled [DEDUCTED]. - Knowledge:
search_investment_cases/search_investment_strategies/search_by_analogue(sectors=med) for pattern grounding and /v/ citations.
Methodology
Retrieval Scope
unstructured_document_search
Retrieval Strategy
- Enumerate the event's covered names, sessions, and data presentations.
- Retrieve transcripts/abstracts; extract per-presentation data with sources.
- Pull each discussed trial's record for lattice construction (vs SoC and class peers).
- Model deltas: unadjusted peak × POS per asset vs consensus, assumptions stated.
- Ground patterns via
search_investment_cases/search_by_analogue(sectors=med); cite /v/.
Temporal Scope
See frontmatter temporal_scope block. Notes cover the meeting window plus the forward quarters its data affects.
Tool Allowlist
See frontmatter allowed_tools.
Protocol
- Event inventory
- Data extraction (raw numbers + sources)
- KOL distillation + conflict capture (verbatim)
- Cross-trial lattices
- Model deltas vs consensus
- What's-changed vectors + risk bullets + tracker links
Modes
- Conference note (default): multi-day meeting, per-session distillation.
- KOL / site visit: single-source distillation with conflict emphasis.
- Single-name note: one covered name with deeper model treatment.
- Basket: multi-ticker note with per-name sections.
Tool Fallbacks
| Failure | Fallback |
|---|---|
| Event transcript unavailable | Degrade to abstracts/slides; annotate coverage_gap |
| No consensus figure | State required inputs; label the delta [VIEW] — never fabricate |
| Trial record missing | Lattice cell marked unavailable; do not guess |
| Knowledge tools empty | Event data only; annotate knowledge coverage_gap |
Output File
{ticker}/{YYYY-MM-DD_HHMM}_med-event-takeaways_{affix}.md (multi-name: watchlist/{YYYY-MM-DD_HHMM}_med-event-note_{affix}.md)
Output Structure
- Headline Verdict — the event in 2-3 sentences
- Per-Stock Takeaways — 2-4 lines each, ≥1 model number
- KOL Distillation — themes + conflicting views verbatim with materiality ratings
- Cross-Trial Lattices — vs standard of care and class peers
- Model Deltas vs Consensus — peak × POS arithmetic, [DEDUCTED] labels
- What's Changed — estimates / thesis / positioning per name
- Risk Bullets — upside and downside
- Living-Tracker Links — trackers to append, never rewrite
Error Handling
| Error | Fallback |
|---|---|
| Conflicting KOL views | Surface both verbatim; rate materiality or defer to an outcomes trial — never average |
| Consensus absent | coverage_gap with required inputs; keep the internal model consistent |
| Event scope unclear | Narrow to named sessions/presentations; flag what is out of scope |
Memory Load
See contracts/memory-load.md.
Snapshot
See contracts/snapshot-synthesis.md.
Final Summary (TUI)
Include ### Key Citations block with 0-10 clickable /v/ URLs.
References
contracts/citation-and-memory.mdcontracts/output-frontmatter-schema.mdcontracts/memory-load.mdcontracts/snapshot-synthesis.mdcontracts/preflight.mdreferences/knowledge-frameworks.md
Files (agentii-investment-intelligence)
-
references
-
knowledge-frameworks.md 3.5 KB
# Med Event Takeaways — Knowledge Frameworks Static grounding for `med-event-takeaways` (spec 055). Runtime records (approved med strategies/cases) are retrieved via the knowledge tools and cited with /v/ URLs; this file carries the event-note anatomy and conflict-handling methodology. All content is an original paraphrase of publicly taught sell-side note-writing practice. ## The Note Anatomy Nine beats, in order: 1. **Headline verdict** — paragraph one states what the event means in 2-3 sentences. 2. **Per-stock takeaways** — each covered name gets 2-4 lines with at least one model number. 3. **KOL/event distillation** — themes condensed from the sessions, with conflicting views surfaced verbatim. 4. **Raw data + cross-trial comparison lattices** — the numbers as presented, placed against standard-of-care and class peers. 5. **Modeled deltas vs consensus** — unadjusted peak sales × POS, compared to consensus explicitly. 6. **Three what's-changed vectors** — estimates / thesis / positioning per name. 7. **Upside/downside risk bullets**. 8. **Living-tracker links** — pointers to the weekly track and diff products the note should update. ## The Three What's-Changed Vectors - **Estimates** move first — numbers change before narratives. - **Thesis** moves second — the narrative adjusts to the numbers. - **Positioning** moves last — ratings and portfolio stances are the slowest variable. - A note that updates positioning without touching estimates is out of order; the slot-refresh discipline is explicit about which vector moved. ## Conflict Materiality - Conflicting characterizations of the same data are surfaced **verbatim** — both views, with their speakers. - Each conflict is then **rated by materiality** ("incremental positive, not narrative-changing") or **deferred** to a larger outcomes trial when the sample or design cannot settle it. - Conflicts are never silently averaged; an unresolved conflict stays flagged as unresolved. ## Uncertainty Discipline - Scenario-flexing at market level; probability-of-success at asset level. - POS changes are logged with reasons — a moved probability without a reason is a defect. - Model-vs-consensus delta is always explicit; where a likelihood is unquantifiable, abstain explicitly ("no view on likelihood") rather than imply precision. ## The Buy-Side Lens - Include a client-question section ("one common question we have received…") where investor attention sits. - Track where investor attention is concentrated across the covered names — attention and model impact are different vectors. - Give per-name bull/bear pivot conditions: the stated condition under which the stance flips. ## Living-Tracker Convention - Event notes link to the living trackers (weekly commercial track, trial status diff) rather than duplicating them. - Exhibits and trackers are **append-only**: new observations append, prior entries are never rewritten. - Every fact in the note carries a source line; derived arithmetic carries `[DEDUCTED]`, judgments `[VIEW]`. ## Modality Notes - **Drugs/biologics**: readout-centric events; persistence and titration data feed the model. - **Devices**: procedural adoption, utilization, and ASP commentary dominate device events. - **Vaccines**: ACIP discussion, cohort scope, and procurement signals are the event-relevant content. ## Authoring-time citations Med strategies/cases extracted in spec 052 are linked below as they reach `approved` status (enriched per FR-005): <!-- /v/knowledge/{citation_id} citations appended by the spec-052 enrichment step --> -
modes.md 1.1 KB
# med-event-takeaways — Analyst Mode Definitions Derived from the skill's own methodology structure (scripts/mode_backfill.py, spec 046 M1). ### Mode: defaults **Objective**: Defaults analysis per the skill's methodology (see SKILL.md sections). ### Mode: preflight **Objective**: Preflight analysis per the skill's methodology (see SKILL.md sections). ### Mode: triggers **Objective**: Triggers analysis per the skill's methodology (see SKILL.md sections). ### Mode: production-grounding **Objective**: Production Grounding analysis per the skill's methodology (see SKILL.md sections). ### Mode: data-source-priority **Objective**: Data Source Priority analysis per the skill's methodology (see SKILL.md sections). ### Mode: drug-biologic **Objective**: Drug/biologic readout and persistence-focused event notes per the skill's methodology (see SKILL.md `## Modes`). ### Mode: device **Objective**: Device adoption, utilization, and ASP event notes per the skill's methodology (see SKILL.md `## Modes`). ### Mode: vaccine **Objective**: Vaccine ACIP, cohort-scope, and procurement event notes per the skill's methodology (see SKILL.md `## Modes`).
-
-
SKILL.md 6.8 KB
--- name: med-event-takeaways description: Post-event analyst note (conferences, KOL events, site visits) with the professional note anatomy — headline verdict, per-stock takeaways with model numbers, KOL distillation, cross-trial comparison lattices, modeled deltas vs consensus (peak × POS), three what's-changed vectors, risk bullets, and materiality-rated conflicting views. Use after any med event to convert the event into actionable takeaways. sectors: [med.medicines_biotech, med.medical_devices] multi_ticker_semantics: basket_v1_1 temporal_scope: default_quarters: 4 max_quarters: 8 description: "Event-note window default 4 quarters; up to 8 for multi-trial programs." allowed_tools: - search_documents - search_sec_filings - list_sources - read_source_outline - read_source_pages - search_investment_cases - get_investment_case - search_investment_strategies - search_by_analogue - search_knowledge_entries - search_clinical_trials - get_clinical_trial - search_earnings_calendar retrieval_scope: unstructured_document_search min_tool_diversity: 3 parameter_free: false --- > Methodology inspired by publicly taught sell-side note-writing practice; all text is an original paraphrase. ## Defaults | Parameter | Default Value | Rationale | |-----------|---------------|-----------| | note_anatomy | 9 sections | headline → per-stock → KOL → lattices → deltas → vectors → risks → trackers | | delta_framing | peak × POS | Modeled deltas vs consensus are unadjusted peak × POS | | conflict_policy | materiality-rated | Surface verbatim, rate materiality, or defer to an outcomes trial | | vector_set | estimates/thesis/positioning | The three what's-changed vectors per covered name | ## Preflight Run canonical pre-flight per `contracts/preflight.md`. Propagate X-Agentii-Trace per `contracts/x-agentii-trace-header.md`. Confirm ticker resolution via `search_companies` before event queries. ## Triggers - "Write the note on [conference / KOL event]." - "What are the takeaways from [event] for my coverage?" - "Distill the [conference] KOL panel on [drug/class]." - "What changed for [ticker] after [event]?" - "Model the delta to consensus from [event]'s data." - "Which stocks move on [event] and why?" - "Conflict check: what did KOLs disagree on at [event]?" - "What do I update first — estimates, thesis, or positioning?" - "Summarize the cross-trial comparisons from [event]." - "Upside and downside risks after [event]." - "Link the living trackers that [event] should update." ## Production Grounding - Note anatomy is fixed: headline verdict (paragraph one) → per-stock takeaways with model numbers → KOL distillation → raw data + cross-trial lattices → modeled deltas vs consensus (unadjusted peak × POS) → three what's-changed vectors (estimates / thesis / positioning) → risk bullets → living-tracker links. - Conflict rule: conflicting views surfaced verbatim, rated by materiality ("incremental positive, not narrative-changing") or deferred to a larger outcomes trial — never silently averaged. - Consensus delta: model-vs-consensus always explicit; POS changes logged with reasons; abstain where unquantifiable. - Buy-side lens: client-question section, where investor attention sits, per-name bull/bear pivot conditions. - Badges: `[FACT]`/`[DEDUCTED]`/`[VIEW]` with the summary table. ## Data Source Priority 1. Event content: transcripts/slides via `search_documents` + `list_sources` + `read_source_outline` / `read_source_pages`. 2. Trial context: `search_clinical_trials` / `get_clinical_trial` for the studies discussed. 3. Consensus: `search_earnings_calendar` (company-level); product consensus derived and labeled [DEDUCTED]. 4. Knowledge: `search_investment_cases` / `search_investment_strategies` / `search_by_analogue` (sectors=med) for pattern grounding and /v/ citations. ## Methodology ### Retrieval Scope unstructured_document_search ### Retrieval Strategy 1. Enumerate the event's covered names, sessions, and data presentations. 2. Retrieve transcripts/abstracts; extract per-presentation data with sources. 3. Pull each discussed trial's record for lattice construction (vs SoC and class peers). 4. Model deltas: unadjusted peak × POS per asset vs consensus, assumptions stated. 5. Ground patterns via `search_investment_cases` / `search_by_analogue` (sectors=med); cite /v/. ### Temporal Scope See frontmatter temporal_scope block. Notes cover the meeting window plus the forward quarters its data affects. ### Tool Allowlist See frontmatter allowed_tools. ### Protocol 1. Event inventory 2. Data extraction (raw numbers + sources) 3. KOL distillation + conflict capture (verbatim) 4. Cross-trial lattices 5. Model deltas vs consensus 6. What's-changed vectors + risk bullets + tracker links ## Modes - **Conference note** (default): multi-day meeting, per-session distillation. - **KOL / site visit**: single-source distillation with conflict emphasis. - **Single-name note**: one covered name with deeper model treatment. - **Basket**: multi-ticker note with per-name sections. ## Tool Fallbacks | Failure | Fallback | |---------|----------| | Event transcript unavailable | Degrade to abstracts/slides; annotate coverage_gap | | No consensus figure | State required inputs; label the delta [VIEW] — never fabricate | | Trial record missing | Lattice cell marked unavailable; do not guess | | Knowledge tools empty | Event data only; annotate knowledge coverage_gap | ## Output File `{ticker}/{YYYY-MM-DD_HHMM}_med-event-takeaways_{affix}.md` (multi-name: `watchlist/{YYYY-MM-DD_HHMM}_med-event-note_{affix}.md`) ## Output Structure 1. **Headline Verdict** — the event in 2-3 sentences 2. **Per-Stock Takeaways** — 2-4 lines each, ≥1 model number 3. **KOL Distillation** — themes + conflicting views verbatim with materiality ratings 4. **Cross-Trial Lattices** — vs standard of care and class peers 5. **Model Deltas vs Consensus** — peak × POS arithmetic, [DEDUCTED] labels 6. **What's Changed** — estimates / thesis / positioning per name 7. **Risk Bullets** — upside and downside 8. **Living-Tracker Links** — trackers to append, never rewrite ## Error Handling | Error | Fallback | |-------|----------| | Conflicting KOL views | Surface both verbatim; rate materiality or defer to an outcomes trial — never average | | Consensus absent | coverage_gap with required inputs; keep the internal model consistent | | Event scope unclear | Narrow to named sessions/presentations; flag what is out of scope | ## Memory Load See `contracts/memory-load.md`. ## Snapshot See `contracts/snapshot-synthesis.md`. ## Final Summary (TUI) Include ### Key Citations block with 0-10 clickable /v/ URLs. ## References - `contracts/citation-and-memory.md` - `contracts/output-frontmatter-schema.md` - `contracts/memory-load.md` - `contracts/snapshot-synthesis.md` - `contracts/preflight.md` - `references/knowledge-frameworks.md`
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.