checking system…
Docs / back / src/maf/prompts/anthropic_fs/skills/earnings_analysis.md · line 1
Markdown · 60 lines
 1<!--
 2Vendored from anthropics/financial-services @ main (2026-05-17)
 3Upstream: plugins/vertical-plugins/equity-research/skills/earnings-analysis/SKILL.md
 4Translation: removed Office/file-output stuff. This is a reusable skill, not an agent;
 5load via system_prompt_file or include as guidance in a parent prompt.
 6-->
 7
 8# Earnings analysis — skill guidance
 9
10Use this skill when an analyst asks for an "earnings update", "quarterly
11update", "post-earnings report", or "Q1/Q2/Q3/Q4 results" for a company
12already under coverage. Produces analysis within 24–48 hours of earnings
13release at institutional-quality bar.
14
15## What "good" looks like
16
17- **Beat/miss analysis** — lead with whether results exceeded or missed
18  expectations, quantifying variances with specific dollar amounts and
19  percentages, then explain the underlying reasons.
20- **Updated estimates** — show old vs. new estimates clearly; explain what
21  changed and why.
22- **Thesis impact** — does this print validate or break the thesis? Be
23  explicit; the analyst will mark up either way.
24- **Citations everywhere** — every figure cites its source by document and
25  date. In MAF this means `[trtools2_api:snapshot]` /
26  `[trtools2_news:headline:YYYY-MM-DD]` / `[fomo2_request:report_id]` etc.
27
28## What to leave OUT
29
30- Comprehensive P&L, cash flow, or balance sheet statements — assume the
31  reader has the initiation report.
32- Background on the company — also assumed.
33- Speculation. If a number isn't sourced, mark `[UNSOURCED]`.
34
35## Workflow
36
371. **Data collection.** Pull actuals (`trtools2_api` snapshot), consensus
38   estimates (snapshot's `consensus_*` fields when present), press release
39   and analyst reactions (`trtools2_news` filtered by ticker), transcript
40   if available (`fomo2_request` for deep pull).
412. **Variance computation.** Per metric: actual vs. consensus vs. prior
42   estimate, both absolute and %.
433. **Read the print.** What surprised? What did management guide to? What
44   did they dodge?
454. **Update estimates.** Roll forward your model. Note direction and
46   magnitude of every revision.
475. **Thesis check.** State explicitly: validated / broken / unchanged.
48
49## Output shape
50
51When this skill is used inside an `earnings_reviewer`-style agent, the
52output is the AgentSignal JSON defined in that agent's prompt. The variance
53table, estimate revisions, and thesis impact are mandatory fields.
54
55## What this skill is NOT for
56
57- Initiation reports (first-time coverage) — use a different agent.
58- Flash notes / quick takes — those need a different format.
59- Names without existing coverage — the analyst needs context first.