3 Statement Agent Skill

Preflight

Run the canonical pre-flight sequence — MCP health probe, ticker resolution, workspace style.md override, memory load, and coverage check. See contracts/preflight.md.

Include the X-Agentii-Trace header on every tool call per contracts/x-agentii-trace-header.md.

Triggers

• analyze 3 statement model
• run 3 statement model analysis
• produce 3 statement model report
• 3 statement model breakdown
• 3 statement model deep dive
• build a 3 statement model
• assess 3 statement model
• quantify 3 statement model
• compare 3 statement model across peers
• review 3 statement model for
• generate 3 statement model on
• 3 statement model for investment decision

Defaults

| Parameter | Default | Notes |

|---|---|---|

| lookback_years | 3 | Historical data window |

| include_peers | false | Whether to surface a peer comparison block |

Methodology

Retrieval Scope

This skill performs unstructured document search at scale across SEC filings (10-K, 10-Q, 8-K). The three-layer agent-use-ready retrieval protocol (Document Discovery → Page Map → Deep Read) applies to all unstructured document search at scale.

Retrieval Strategy

See contracts/retrieval.md for the canonical decision tree; skill-specific retrieval detail is in references/methodology.md.

Temporal Scope

Default: 12 fiscal quarters (max 20). Financial modeling: trailing 12 quarters (3 fiscal years) for long-range projection inputs.

Tool Allowlist

See frontmatter allowed_tools.

Protocol

Step-by-step execution detail is in references/methodology.md.

Deliverable Chain

Inputs → Build → Validate → Output → Next

• Inputs: resolved ticker + 3 years of search_xbrl_facts (Income Statement, Balance Sheet, Cash Flow) + get_statement_structure for presentation tree.
• Build: write a self-contained Python script using openpyxl that creates the 3-statement workbook per ## Output Structure. Execute via Bash: python3 script.py. Verify the .xlsx exists at the output path. If import openpyxl fails, fall back to .md summary with data_availability: degraded (see contracts/office-tooling.md).
• Validate: run LibreOffice recalc; audit cross-statement checks (BS balances, CF ties to BS, IS flows to CF) per ## Validation Gates.
• Output: write the artifact path per ## Output File.
• Next: append to agentii.md; hand off to a downstream pitch/review skill if requested.

Validation Gates

• balance sheet balance: Assets = Liabilities + Equity within 1% tolerance. *If failed*: If unbalanced > 1%: refuse delivery, report imbalance amount.
• cash flow tie-out: CF ending cash = BS cash for current period. *If failed*: If mismatched: refuse delivery, report discrepancy.
• forecast years: exactly 5 historical + 5 forecast years. *If failed*: If < 5+5: flag in Coverage Gaps, proceed with available data.
• calculation arc cross-validation : cross-statement balancing verified against gold.xbrl_calculations weights — each parent concept's value equals the weighted sum of its children per the XBRL calculation linkbase (e.g., Assets = +1.0 × CurrentAssets + 1.0 × NoncurrentAssets, NetIncomeLoss = +1.0 × Revenues - 1.0 × OperatingExpenses + ...). Call get_statement_structure/{ticker}?statement_type=<type>&include_calculations=true to retrieve the weighted parent-child relationships. Flag discrepancies ≥1% of parent concept value as audit findings. *If failed*: If any material discrepancy (≥1% of parent value): flag in audit findings, refuse delivery for discrepancies ≥5%.

• tool diversity: distinct MCP tools used in this invocation >= min_tool_diversity (5). *If failed*: flag as depth-insufficient in Coverage Gaps, listing which tool categories were unused (structured data / document retrieval / company metadata / earnings calendar / coverage). This gate does NOT block analysis completion — it is a quality signal for your review.

Tool Fallbacks

Per-tool failure modes and fallback actions are tabulated in references/tool-fallbacks.md.

Output File

Write the final deliverable to {ticker}/{YYYY-MM-DD_HHMM}_3-statement_{affix}.md .

Output Structure

The deliverable is a structured markdown report written to the path in ## Output File. Full section-by-section template (headings, tables, and field definitions) lives in references/output-structure.md. Required elements:

• Executive Summary — headline conclusions (≤200 words).
• Core analysis sections — per this skill's methodology and analyst modes.
• Data classification — tag findings [FACT] / [DEDUCTED] / [VIEW] per contracts/snapshot-synthesis.md.
• Coverage Gaps & Citations — inline /v/ citations are PRIMARY (immediately after each fact); the bottom Citations section is a non-duplicative roll-up index.
• Output frontmatter — emit the FR-090 structured block per contracts/output-frontmatter-schema.md.

Citations & memory: follow contracts/citation-and-memory.md — ≥1 citation per 200 words; every material fact, table row, and metric is immediately followed by its inline clickable https://agentii.ai/v/{ticker}/{citation_id}/{N} link; a bottom Citations section provides a non-duplicative roll-up index; the closing TUI reply includes a compact Key Citations list (headline 5–10 facts) of clickable /v/ URLs; and append the run to agentii.md per contracts/agentii-md-schema.md.

Memory & Snapshot

• Memory load (pre-flight): load prior workspace context for the ticker before retrieval — see contracts/memory-load.md.
• Structured output frontmatter: emit the FR-090 block (key_metrics, conclusions, facts_count, deducted_count, views_count, citation_count) per contracts/output-frontmatter-schema.md.
• Snapshot synthesis: after writing the deliverable, update the two-tier snapshot and classify findings as [FACT]/[DEDUCTED]/[VIEW] — see contracts/snapshot-synthesis.md.
• Session archival: record the run under sessions/{YYYY-MM-DD}/ and update sessions/INDEX.md per contracts/session-format.md.

Final Summary (TUI)

End the closing chat reply with a compact Key Citations list (headline 5–10 facts), each a clickable https://agentii.ai/v/{ticker}/{citation_id}/{N} link, so the user can cmd+click straight to the exact SEC page. See contracts/citation-and-memory.md.

Error Handling

| Failure Mode | Detection | Action | User-Facing Message |

|---|---|---|---|

| Missing data | Data API returns empty result set | Widen date range and retry once | "No data available for {ticker} in requested window." |

| Partial data | Data API returns <80% expected records | Proceed with coverage gaps section | "Analysis based on partial data; see Coverage Gaps section." |

| Sector mismatch | Peer sector != target sector | Filter out mismatched peers | "Removed {n} peer(s) due to sector mismatch." |

| Insufficient history | Ticker <3 years on public markets | Downgrade to limited-history profile | "Limited historical data; analysis adjusted accordingly." |

| MCP unreachable | Preflight probe fails | Halt with actionable error | "agentii data plane unreachable; check connection." |