Strategy Chat

Run a multi-turn strategy Q&A workflow for: $ARGUMENTS

Overview

Implementation status: workflow-only
Current backing path: compose analysis, market-brief, technical-scan, fundamental-context, and news-intel
Primary purpose: keep a multi-turn strategy discussion coherent with rigorous artifact reuse and provenance discipline
Research layer: conversational wrapper over descriptive, decision, or execution layers, depending on the turn

Use When

The user asks follow-up questions instead of wanting a single report.
The user wants to explore scenarios, invalidation points, or alternative strategy styles conversationally.
The caller needs continuity over the same symbol, run artifact, or saved analysis rather than restarting the workflow every turn.
The user wants to iterate on a thesis or decision frame through dialogue.

Do Not Use When

The user wants a one-shot concise report. Use market-brief.
The user wants only a direct action or quantity. Use decision-support.
The user wants exact artifact retrieval or export. Use reports.
The user wants broad historical comparison rather than turn-by-turn discussion. Use analysis-history.
The conversation has no clear anchor (symbol, artifact, or prior analysis). Clarify the anchor first.

Inputs

A user question, optionally plus a stock symbol.
Preferably an explicit prior artifact, run, or earlier analysis result.
Optional prior context from earlier turns in the same session.
Optional user assumptions such as capital, position, benchmark, holding horizon, or preferred style.

Execution

Step 1: Anchor the symbol, horizon, and evidence explicitly

Mandatory anchor requirements:

Every substantive turn must have a clear anchor: a specific symbol, a saved artifact, or an explicit recent analysis result
If the anchor is ambiguous (e.g., user says "what about the risks" without specifying which symbol or analysis), ask for clarification before proceeding
If reusing a prior artifact, state which artifact (path, generation time, command) is being used as the anchor
If reusing session context from a prior turn, state what context is being reused

Anchor clarity rules:

Do not assume the user wants to continue discussing the same symbol if they haven't mentioned it in the current turn
Do not assume prior assumptions (capital, holding period, style) still apply if the user's question suggests they may have changed
When in doubt, ask: "Are you still asking about [symbol] based on [artifact/prior turn]?"

Step 2: Reuse prior artifacts instead of recomputing blindly

Artifact reuse discipline:

Prefer saved analysis or market-brief artifacts when answering follow-up questions
State which artifact is being reused: path, generation time, and command
Recompute only when:
- The user explicitly asks for a refresh
- The old evidence is too stale for the question (state the staleness threshold used)
- The user has changed key assumptions that invalidate the prior artifact

Staleness assessment:

If the artifact is more than [N] days old and the user is asking about current positioning, recommend a refresh
If the artifact was generated before a material event (earnings, news, regime change), recommend a refresh
State the staleness assessment explicitly: "The last analysis was generated [X] days ago; recommend refreshing if current positioning is needed"

Step 3: Answer in labeled mode with explicit layer identification

Layer labeling requirements: Each substantive reply must make clear which mode it is operating in:

Descriptive analysis mode: interpreting current market state without action or sizing
Decision-support mode: providing conditional action and sizing guidance with explicit assumptions
Execution-planning mode: designing entry/exit zones and checklists for an accepted view
Retrospective evaluation mode: reviewing prior decisions against outcomes

Mode transition handling:

If the user changes mode (e.g., from asking about current state to asking about position sizing), acknowledge the transition explicitly
State which new assumptions are needed for the new mode
If prior assumptions no longer apply, ask the user to confirm or update them

Step 4: Track assumption changes and invalidation

Assumption tracking:

If the user changes horizon, position context, or thesis assumptions, state whether the prior conclusion still applies
If assumptions have changed materially, recommend regenerating the decision or execution frame rather than stretching stale context
Maintain a running list of active assumptions for the conversation (symbol, horizon, capital, position, style, etc.)

Invalidation monitoring:

If the prior analysis included invalidation conditions, check whether any have been triggered
If invalidation conditions have been met, state that explicitly and recommend reassessment
Do not let an old buy, hold, sizing, or execution view survive unexamined after assumptions change or invalidation conditions trigger

Step 5: Guard against conversational drift and false continuity

Drift prevention:

Continuity is conditional on stable anchor and assumptions, not free-form drift
If the conversation has drifted away from the original anchor, acknowledge that and establish a new anchor
Do not blend multiple symbols, timeframes, or thesis frames into a single confused narrative
If the user asks a question that cannot be answered using the current anchor, say so explicitly

False continuity prevention:

Do not pretend that conversational memory is equivalent to saved artifacts
Do not carry forward conclusions from prior turns if the evidence basis has changed
If the user references "what we discussed earlier" but no artifact exists, clarify what is being referenced

Output Contract

Expected result: one conversational but analysis-grounded response.
Caller-facing delivery standard:
- Mandatory anchor disclosure: name the anchor basis for the current answer (fresh run, saved artifact path, or prior turn context)
- Artifact provenance: if reusing a saved artifact, state path, generation time, and command
- Layer identification: state whether the current reply is descriptive analysis, decision support, or execution planning
- Assumption tracking: identify unresolved assumptions or changed assumptions before presenting a confident conclusion
- Staleness assessment: if reusing old evidence, state how old it is and whether refresh is recommended
- Invalidation monitoring: if prior analysis included invalidation conditions, check whether they have been triggered
- Prefer artifact reuse over regeneration: only recompute when explicitly requested or when staleness/assumption changes require it
Local limitation: no dedicated local chat persistence engine exists. Continuity depends on the host framework and the explicit anchor provided.

Failure Handling

If the symbol or prior context is ambiguous, say so and ask the host framework to clarify through normal interaction.
If supporting data is partial, continue with bounded uncertainty rather than pretending full confidence.
If the user changes the thesis materially, recommend a refresh through market-brief or analysis instead of stretching stale context.
If the anchor is lost (e.g., user switches symbols without stating it), ask for clarification rather than guessing.
If assumptions have changed but the user hasn't acknowledged it, surface the change and ask whether prior conclusions still apply.

Key Rules

Continuity is conditional, not free-form drift. Stable anchor and assumptions are required.
Prefer persisted artifacts or explicit recent outputs over fuzzy conversational memory.
Do not let an old buy, hold, sizing, or execution view survive unexamined after assumptions change.
Keep the answer strategic rather than collapsing into generic chat.
Anchor clarity is mandatory. Every turn must have a clear symbol, artifact, or prior analysis basis.
Layer identification is mandatory. State which mode (descriptive, decision, execution) the current turn is operating in.
Assumption tracking is mandatory. Maintain and surface the active assumption set for the conversation.
Staleness assessment is required when reusing artifacts. State how old the evidence is and whether refresh is recommended.
Invalidation monitoring is required. Check whether prior invalidation conditions have been triggered.

Composition

Usually builds on analysis, market-brief, technical-scan, fundamental-context, and news-intel.
Should hand off to reports, analysis-history, decision-support, or strategy-design when the user narrows the task into a cleaner adjacent skill.
Should route to backtest-evaluator when the user wants to evaluate a prior decision against outcomes.

strategy-chat