bichengwang/financial-agent/equity-publish-gate

Claude Code and agent skill by bichengwang/financial-agent

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Checks a daily equity research package before it is published - required artifacts on disk, 15_predictions.json schema and units, formula conformance (70% CI, mu calibration band, Adj Score trace, VaR/CVaR), core ETF forecasts, and a replay of the run status from the ledger. Use before moving a daily investment run to PUBLISHED, when a manifest checklist is being written, or when auditing prediction ledgers for drift across runs and models.

Install

npx vskill@latest install bichengwang/financial-agent/equity-publish-gate

SKILL.md

Equity publish gate

The harness, not this skill, owns the rules: the checks live in src/financial_agent/harness/ and the numbers in its equity_policy.toml. This skill tells you when to run them and how to act on the result.

Before publishing a run

  1. Write every artifact first. Never write the manifest's artifact checklist

before the files exist; the gate reads the directory.

  1. Run the gate on the package:

``bash PYTHONPATH=src python3 -m financial_agent.harness gate agents/equity/output/{model}-{YYYY-MM-DD} ``

  1. Exit status 0 and publish gate: PASS means the package may move to

PUBLISHED. Copy the status replay and content hash lines into 00_run_manifest.md (the hash pins exactly what was published); if the replay disagrees with your status, explain why in 08_risk_review.md (the replay assumes all Required inputs are grounded and knows nothing about integrity halts). The replay applies the NYSE calendar: a market holiday publishes REVIEW_ONLY; a weekend keeps the computed status until a weekend rule is set in the policy.

  1. On GATE FAIL, fix the cause and regenerate the artifact. Do not edit the

policy file or loosen a check to get a pass. A failure you cannot fix makes the run HALTED, not published.

To generate the manifest's artifact checklist, gate result, status replay, and content hash from the finished directory, run PYTHONPATH=src python3 -m financial_agent.harness manifest agents/equity/output/{model}-{YYYY-MM-DD} and paste its output into 00_run_manifest.md. Never type the checklist by hand.

A ledger that declares "schema_version": 1 is also held to ledger schema v1 (one run_status, split Kelly fields, decimals, canonical settlement keys). PYTHONPATH=src python3 -m financial_agent.harness schema agents/equity/output/{model}-{YYYY-MM-DD} lists its v1 findings; schema --json-schema prints the shape.

Reading the output

  • error: breaks an explicit rule (missing artifact, enum, CI or mu-band

formula, Adj Score trace). Blocks publication.

  • drift: a unit or meaning the rules leave implicit (VaR stored as percent,

kelly_025 capped in one run and uncapped in the next). Report it in 13_evolution_log.md as a schema-clarity observation. Do not "fix" history.

  • GO unreachable: threshold N ...: no name can pass that evidence threshold

with the factor families that can score. This is structural, not a property of today's names: cite it as the reason for NO_TRADE rather than re-deriving it, and do not count it as a new finding each run.

Auditing history

python -m financial_agent.harness clones --since YYYY-MM-DD lists artifacts that are byte-identical in more than one package. A clone means an analysis file was copied instead of produced; explain or regenerate it.

python -m financial_agent.harness replay --since YYYY-MM-DD recomputes every published status from its ledger and lists the packages that disagree.

bash
PYTHONPATH=src python3 -m financial_agent.harness audit --output-dir agents/equity/output

Historical packages are immutable: never rewrite a prior 15_predictions.json. Use the audit table as evidence for a Track B schema proposal instead.

Security

  • Security scanPassed, 98 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
45
Repository forks
9
7-day trend score
0

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/bichengwang/financial-agent/equity-publish-gate/badge)](https://verified-skill.com/skills/bichengwang/financial-agent/equity-publish-gate)

Something wrong with this skill or its page? Tell us.