agentleFS
KCNyuKCNyu/clawock/skills/earnings-review/SKILL.md

earnings-review

Skill23 starsChanged 3 months agoMITView on GitHub

Manual, event-driven earnings review for a US or HK holding, backed by first-party filings. Use when a company reports, when a management promise comes due, or when a thesis review needs primary-source numbers. Produces a structured artifact under memory/earnings/<TICKER>/<period>.json whose earnings-quality math, promise ledger, and thesis evidence are computed by the packaged clawock earnings command — not by prose. Never places trades and never changes thesis state.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Never assert a computed number in prose.

    line 25
  2. Very specific:

    Every published number needs two independent sources in provenance (see clawock.provenance).

    line 71
  3. Distinctive:

    No persona scoring and no aggregate "master score" — a red-line or integrity failure is never averaged away by good numbers elsewhere.

    line 104
  4. Distinctive:

    No trade from a single earnings classification.

    line 106

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Earnings Review

Event-driven, never scheduled. Run this when an issuer reports, when a commitment comes due, or before a thesis review that needs first-party numbers. It is not part of any cron path.

The output is not a report. It is memory/earnings/<TICKER>/<period>.json, and the Markdown you write afterwards is a rendering of that file.

What the code owns and what you own

Owned by clawock earningsOwned by you
Source grade A/B/C and whether footnote claims are allowedReading the filing and paraphrasing what it says
Cash conversion, FCF, working-capital gaps, dilution, SBC share, marginsChoosing which segments and footnotes matter
Guidance beat / inline / missRecording the commitment and its measurable target
Promise roll-forward to met / partial / missed / not_due / unverifiableLocating the document that proves the result
The release gate over the provenance manifestDeciding whether the period changes the thesis

Never assert a computed number in prose. Run the script and quote its output.

Step 1 — collect first-party documents

Source order is not negotiable:

  • US — SEC filing (10-K/10-Q/8-K) or issuer IR first; clawock filings supplies the structured XBRL numbers used to verify them.
  • HK — HKEX announcement or issuer IR first; clawock fundamentals (Eastmoney) is a secondary structured source, never a substitute for a footnote.
  • A third-party summary may only fill a gap, and it lowers the grade.

Record each document once in documents[] with its source_class, a stable locator, retrieved_at, and whether it covers_period. Store a short paraphrase and the locator — never long transcript passages.

Grades are mechanical: A needs a covering primary document plus a structured dataset; B has the dataset but no covering primary document; C is third-party only. B/C disable every footnote claim, and the validator enforces that. A low grade describes the sources, not the company.

Step 2 — build the comparable history

At least four comparable quarters (or four half-years, or three annual periods). basis, currency and unit must be identical across every comparable — a GAAP/non-GAAP or currency switch mid-history is a validation error, not a caveat. revenue, net_income and ocf are mandatory per period; anything else you omit becomes an explicit unavailable with a reason instead of a number.

Step 3 — promises and capital allocation

Carry the previous period's ledger forward; a promise may never be dropped, and met/partial/missed are terminal.

clawock earnings promises \
  memory/earnings/TICKER/<previous>.json memory/earnings/TICKER/<current>.json

An overdue promise with no reported result becomes missed when this reporting period covers its due date, and unverifiable when the due date has merely passed with no covering report yet. Log buybacks, dividends, M&A, leverage changes, divestitures and new-business spend in capital_allocation[].

Step 4 — provenance and release

Every published number needs two independent sources in provenance (see clawock.provenance). The release gate refuses the artifact when any number is single-sourced or the two sources disagree beyond tolerance.

clawock earnings validate memory/earnings/TICKER/<period>.json
clawock earnings review   memory/earnings/TICKER/<period>.json

review prints the source grade, the provenance verdict, the quality metrics and anomaly_flags. A non-zero exit means the artifact is not releasable — fix the artifact, do not narrate around it.

Step 5 — hand evidence to the thesis, not a verdict

clawock earnings thesis-evidence memory/earnings/TICKER/<period>.json

This emits evidence[] rows in the thesis registry's shape plus dimension_suggestions. It deliberately carries no thesis state. To change a thesis, append these evidence rows to a new version of memory/theses/<thesis-id>.json and run the registry's own drift evaluator:

clawock thesis drift old.json new.json

The registry re-checks freshness itself and will reject a dimension change that leans on evidence older than the last check.

Hard limits

  • No persona scoring and no aggregate "master score" — a red-line or integrity failure is never averaged away by good numbers elsewhere.
  • No trade from a single earnings classification. Sizing stays with the existing decision and risk contracts.
  • A price move or a headline is not a filing fact.
  • Missing documents lower the grade. They never justify a fabricated number.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.