Daily Stock Analysis

Perform market-aware, evidence-based daily stock analysis with prediction, next-run review, rolling accuracy tracking, and a structured self-evolution mechanism that updates future assumptions from observed forecast errors.

Hard Rules

1. 1. Read and write files only under working_directory.
2. Save new reports only to:

• - INLINECODE1

1. 3. Use filename:

• - INLINECODE2

1. 4. If same ticker/day file exists, ask user:

• - overwrite or new_version (-v2, -v3, ...)
• For unattended runs, default to INLINECODE7

1. 5. Always review history before new prediction.
2. Limit history read count to control token usage:

• - Script mode: max 5 files (default)
• Compatibility mode: max 3 files

Required Scripts (Use First)

1. 1. Plan output path + collect history:

CODEBLOCK0

1. 2. Compute rolling accuracy from existing reports:

CODEBLOCK1

1. 3. Optional: migrate legacy files after explicit user confirmation:

CODEBLOCK2

Compatibility Mode (No Python / Small Model)

If Python scripts are unavailable or model capability is limited, switch to minimal mode:

1. 1. Read at most 3 recent reports for the same ticker.
2. Use only a minimal source set:

• - one official disclosure source
• one reliable market data source (Yahoo Finance acceptable)

1. 3. Output concise result only:

• - recommendation
• INLINECODE8
• prior review (prev_pred_close_t1, prev_actual_close_t1, AE, APE) if available
• one INLINECODE13

1. 4. Save report with same filename rules in canonical reports directory.

See references/minimal_mode.md.

Minimal Run Protocol

1. 1. Resolve ticker/exchange/market (ask if ambiguous).
2. Run report_manager.py plan.
3. Read history_files returned by script.
4. If legacy_files exist, list all absolute paths and ask whether to migrate.
5. Gather data using references/sources.md + references/search_queries.md.
6. Run calc_accuracy.py for consistent metrics.
7. Render report using references/report_template.md.
8. Save to selected_output_file returned by report_manager.py.

Required Output Fields

Must include:

• - INLINECODE24
• INLINECODE25
• INLINECODE26
• INLINECODE27
• INLINECODE28, INLINECODE29
• rolling strict/loose accuracy fields
• INLINECODE30

Self-Improvement (Required)

Each run must include 1-3 concrete improvement_actions from recent misses and use them in the next run.
Do not skip this step.

Scheduling Recommendation

Recommend users set this as a weekday recurring task (for example 10:00 local time) to keep prediction-review windows continuous.

References

Default:

• - INLINECODE32
• INLINECODE33
• INLINECODE34
• INLINECODE35
• INLINECODE36
• INLINECODE37
• INLINECODE38

Deep-dive only (full_report mode):

• - INLINECODE40
• INLINECODE41
• INLINECODE42

Compliance

Always append:

"This content is for research and informational purposes only and does not constitute investment advice or a return guarantee. Markets are risky; invest with caution."