| name | stockvaluation-io |
| description | Use whenever the prompt mentions stockvaluation.io, StockValuation.io, stockvaluation, the local valuation MCP, or asks to value a public company with the StockValuation workflow. Runs local MCP valuation with server-enforced gates and anchored defaults, stops for evidence review, asks guided valuation-refinement questions, and builds the report with the bundled report builder and deterministic prose linter. |
| version | 3.0.0-workflow-consistency |
| homepage | https://github.com/stockvaluation-io/stockvaluation_io |
StockValuation.io Agent Valuation
Use this skill whenever the prompt mentions stockvaluation.io, stockvaluation, the local valuation MCP, or asks you to value a public company, critique DCF assumptions, build scenarios, or troubleshoot the local service. The deterministic valuation math comes from local MCP tools; you orchestrate the workflow, gather evidence, and supply judgment. Educational use only; this is not financial advice ({baseDir}/references/no-advice-policy.md).
A plain request such as "value COMPANY using stockvaluation.io" is the default full researched valuation flow — not a quick valuation and not a one-shot report request. Never infer an evidence-review or guided-refinement bypass from ordinary phrasing; bypass only when the user explicitly says quick, no questions, skip questions, one-shot report, automation, or smoke-test, and record the bypass through gate_records.
Server-Enforced Gates
The MCP server tracks each run. Always pass the run_id returned by stockvaluation.researched_baseline, stockvaluation.value_ticker, or stockvaluation.extract_prospectus to every downstream call. On tracked runs the server refuses out-of-order calls:
- Scenario-bearing valuation before the evidence-review gate is recorded →
GATE_NOT_CLEARED.
- Guided-flow recalculation before answers are applied or an explicit bypass is recorded →
GATE_NOT_CLEARED.
- A numeric driver value that is neither a server-computed anchor nor a recorded guided answer with
source=user_input → UNANCHORED_SCENARIO_VALUE or UNVERIFIED_USER_INPUT. Never author scenario numbers yourself.
- While material drivers are unresolved, scenario calls return a low/high
valuationRange; lead with that range and name the unresolved drivers. A point estimate exists only when every material driver is pinned.
Default Workflow (ticker)
stockvaluation.health, then stockvaluation.researched_baseline (keep stockvaluation.value_ticker mechanical). Keep the returned run_id.- Handle
sourceQualityGate stops per {baseDir}/references/workflow.md.
- Run segment discovery (
{baseDir}/references/segments.md) before treating anything as the researched mechanical baseline; use a segment-aware mechanical baseline when a validated package exists.
- Build the evidence packet with research subagents and classify driver-specific evidence (
{baseDir}/references/evidence-and-judgment.md).
- Stop at the evidence review gate (
{baseDir}/references/workflow.md); record the outcome via gate_records.
- Run baseline plausibility and the method checks (
{baseDir}/references/valuation-method.md), produce assumption_judgment, and auto-recalculate once via stockvaluation.recalculate with request_policy.mode = "autonomous_researched" when a governed change is supported. Do not hand-compute valuation math.
- Run guided refinement:
stockvaluation.plan_guided_questions with the run_id (the server attaches anchored low/base/high choices), ask one question at a time, then stockvaluation.apply_guided_answers and one final recalculation with request_policy.mode = "user_refined_scenario". See {baseDir}/references/workflow.md.
- Build the report with
{baseDir}/scripts/build_report.py per {baseDir}/references/report.md; the prose linter runs inside it and the generated index.html opens automatically unless disabled with --no-open or STOCKVALUATION_OPEN_REPORT=0. Return the clickable index.html link.
Prospectus Workflow
For a SEC EDGAR HTML prospectus URL: stockvaluation.extract_prospectus, stop at the extraction review card, then stockvaluation.value_prospectus with review_reference and review_status=reviewed only after user review. Continue into the normal evidence review and guided refinement; the final scenario call carries prospectusScenarioCandidate.scenario. Details and basis-status rules: {baseDir}/references/workflow.md.
Judgment Rules
- Treat MCP JSON as the source of truth; never invent missing service fields, financial data, or scenario math.
- Evidence must name a driver and a fact; generic source presence ("10-K found") is not evidence. User answers and approvals are user judgment, not evidence.
- Market-implied and priced-in data are report-only diagnostics, never evidence or autonomous changes.
- Financial firms, private companies, and insufficient-data cases stop cleanly (
{baseDir}/references/valuation-method.md); use stockvaluation.explain_failure and do not invent a valuation.
- Accounting topics follow
{baseDir}/references/accounting-and-claims.md: explain returned values; apply R&D capitalization only through the governed source-backed R&D path.
- Keep MCP arguments compact (
{baseDir}/references/mcp-tools.md).
References
{baseDir}/references/mcp-tools.md — tool reference, run tracking, anchors, failure recovery{baseDir}/references/workflow.md — gates, guided refinement, prospectus mode{baseDir}/references/evidence-and-judgment.md — evidence rules, assumption judgment, plausibility{baseDir}/references/valuation-method.md — method checks, coverage boundary, stop rules{baseDir}/references/segments.md — segment discovery and quality{baseDir}/references/accounting-and-claims.md — accounting statuses and claims{baseDir}/references/report.md — report builder, prose linter, HTML artifact{baseDir}/references/no-advice-policy.md — policy (verbatim, controlling)
