finance-research-agent: Skill for Claude Code

Management credibility

Trust is earned by delivery. This skill compares what management PROMISED against what the company REPORTED, scores the gap with a precise, repeatable methodology (not a vibe), and reads the language around misses for evasiveness patterns — so the user can judge credibility on evidence, not tone alone.

Procedure

1. Resolve the symbol; note the company type — guidance style differs materially (IT services guide revenue-growth ranges and margin bands; manufacturers guide capex and capacity; banks guide credit growth, NIM and credit-cost).

2. Pull both sides — management_guidance (default lookback_periods=6; use 8-10 for a fuller track record). It returns:

   • dated forward-looking / guidance statements retrieved from past concalls (revenue-growth, margin, capex, outlook themes), and
   • the actual reported quarterly results (Sales, OPM %, Net Profit, EPS). If concalls aren't indexed for the company, it says so and still returns the actuals — note the limitation and don't force a credibility verdict off actuals alone.

3. Score each guidance instance with a precise variance band, not a loose "close enough": Variance % = (Actual - Guided) / |Guided| × 100 (for a guided range, compare against the nearer band edge; for a directional guide with no number — e.g. "margins should improve" — score it qualitatively as met/missed against the reported change, and say you did).

   • Beat: actual exceeds guidance by >5%
   • Met: actual within ±5% of guidance
   • Modest miss: actual below guidance by 5-15%
   • Large miss: actual below guidance by >15%, or guidance quietly dropped/revised down without being flagged as such by management Tabulate every instance found — don't cherry-pick the clean ones.

4. Compute the summary scores:

   • Hit rate = (instances scored Met or Beat) / (total instances) over the lookback window — state as e.g. "6/8 quarters (75%)".
   • Directional accuracy vs magnitude accuracy — separate "called the right direction but missed the size" (e.g. guided "modest growth", delivered 2% when they meant closer to 8%) from "called the wrong direction entirely" (guided growth, delivered a decline). The latter is materially worse for credibility than the former.
   • Sandbagging vs over-promising pattern — check whether misses cluster as chronic over-promising (repeated large misses, same direction every time) vs chronic sandbagging (guidance consistently beaten by a wide, suspiciously consistent margin — itself a credibility flag, since it implies guidance is being lowballed rather than genuinely forecast).