Stock Screener

Find credible opportunities and explain the expectations embedded in their prices. Maintain one company-evaluation method for discovered stocks and existing holdings.

Modes And Ownership

• Screen opportunities: read references/opportunity-screening.md, build an auditable candidate funnel, then evaluate finalists with the shared evaluator.
• Evaluate companies: read references/company-evaluation.md completely before rating named stocks. A request from portfolio-analysis uses this mode; it does not trigger discovery beyond the supplied holdings.
• For turnarounds, order books, capacity expansion, recovery or other catalyst-led theses, also read references/catalyst-watchlist-framework.md.

The screener owns company merit and research evidence. Portfolio-analysis applies holdings, fund look-through, taxes and allocation constraints to those findings. When both tasks are requested, share the evidence instead of independently recalculating company value. No separate evaluator skill is required.

Working Method

1. Establish the as-of date, universe, currency, horizon, budget and required return. Preserve explicit exclusions; distinguish hard restrictions from soft preferences. If the hurdle or order size is absent, disclose an illustrative assumption and show sensitivity instead of treating it as the user's mandate.
2. Verify current universe membership, identifiers and price dates. In India, BSE trading groups, index membership, AMFI size bands and F&O eligibility are separate attributes. None alone certifies business quality or manipulation. Use a current official list for a requested universe or exclusion.
3. In discovery mode, maintain counts and reasons through eligibility, quantitative triage and company research. Separate unknown data from rejection. Examine sector-specific value, durable growth and recovery/catalyst paths; a few familiar names cannot establish that a large universe has no opportunities.
4. Evaluate operating quality, normalized per-share earnings, governance, funding, cash conversion and the appropriate valuation lens. Reconcile material differences between reported and normalized earnings with dated sources. Separate firm backlog from optional pipeline and management-adjusted metrics from statutory earnings and cash flows using the relevant references.
5. Model bear, base and bull outcomes. For representative positive EPS, run scripts/project_returns.py from this skill using a temporary JSON outside the source repository. Report price CAGR, cash-flow IRR, terminal-wealth CAGR and dividend assumptions separately. The cash-flow IRR and its discounted entry value are the default comparison with an annual return hurdle.
6. Show growth, margin, exit-valuation and hurdle sensitivity for close calls. Reverse the valuation to expose what the price requires. Keep unsupported recovery or rerating in an upside scenario, rather than adjusting assumptions until a desired stock passes.
7. Rank investment merit before assessing portfolio suitability. An attractive stock can remain shortlisted with a zero suggested allocation because of existing exposure. Use states Discover, Validate, Starter, Scale/Add, Hold, Trim and Exit/Archive; distinguish research rejection from a sell decision on an existing holding. State confidence and evidence that changes it. Keep eligibility exclusions separate: an excluded or unreviewed security has no assessed company state, rather than an Exit/Archive investment judgment.

Decision Discipline

• Verify a claim such as "multi-year low" for the specified index or stock and distinguish price drawdown from valuation and normalized earnings. Index-level P/E is context, not a reason to reject every constituent.
• Use liquidity on all relevant trading venues over a stated multi-week window, with spread/free-float checks and realistic order participation. One day on one exchange and a fixed market-cap floor do not establish investability.
• Investigate pledge amount, denominator, trend, purpose, collateral and financing stress. Do not turn every pledge into a fraud finding or overlook forced-sale risk because earnings look attractive.
• Allow an evidence-backed, capped starter when uncertainty is measurable and survivable. Stronger operating and cash-flow confirmation is normally needed before scaling; waiting for all uncertainty to disappear is not an entry rule.
• Present a ranked watchlist when no immediate buy is validated. Do not manufacture five buys or claim that no opportunity exists in an incompletely researched universe. Disclose missing evidence and the next research step.
• A prior rally or decline is a reason to audit assumptions and timestamps, not proof of the original thesis. Acquisition cost is relevant to taxes, not value.
• A platform's buy consensus or favourable entry indicator is a research lead, not evidence of intrinsic value. Verify what the signal measures, its date and coverage; do not treat technical entry conditions as fundamental cheapness.

Output

For discovery, include:

• dated mandate and universe sources, funnel counts, sector coverage and data gaps;
• an eligibility/research ledger with identifiers, state, reason, dates and links;
• ranked company findings with normalized record, valuation lens, bear/base/bull returns, sensitivity, buy range, confidence, catalyst and invalidators;
• liquidity/order-size limits and governance/funding risks;
• rejected finalists and near misses with their decisive reasons;
• portfolio suitability as a separate column when holdings are known.

For named-company evaluation, return the company findings to the caller without creating an unrelated candidate funnel or allocation plan. Missing primary data means Discover or Validate, not a fabricated high-confidence rating. Evaluate an explicitly requested company even when it is outside the new-buy universe, but flag its eligibility separately; the request is not permission to relax purchase restrictions. Keep research state separate from the caller's decision to hold, add to or sell an existing position.

Keep generated reports, ledgers, input data and downloaded evidence beside a user-supplied input or in a private temporary workspace. They do not belong in this public skill repository. Use ranges for decisions; calculator precision is not forecast precision.