Agent Skills
A public SKILL.md teaches compatible agents when to route money-market, capital-market, China-economy, bank-risk, and market-liquidity research across the four products.
Verified release 0.1.7: GIFT City, forex and gold join the five existing topics in the published package and public MCP router. The helper reads fixed public context; user-specific scenarios remain explicit steps.
Open financial research infrastructure · verified release v0.1.7
Financial Evidence connects terminal and agent workflows to LiquiLens, Undertow, Seiche, and Palimpsest. It routes each question to the relevant public evidence without inventing a universal score, hiding an unavailable source, or requiring an account or API key.
Verified release: v0.1.7 is a signed, independently verified release. Agent Skill, terminal, Codex, Gemini, editor, and bundle installs use the published tag and artifacts, container, Registry record and separately verified Homebrew formula. The Claude self-hosted command follows mutable default-branch source; review its commit before enabling it.
Deterministic topic routing
Use Seiche for money and capital markets, GIFT City, forex and gold context, LiquiLens for covered bank and lender risk, Undertow for market liquidity, and Palimpsest plus Seiche for China-economy evidence. The router preserves separate documents, canonical human scope URLs, source URLs, retrieval clocks, byte counts, content hashes, and explicit error states.
| Question | Product route | Evidence boundary |
|---|---|---|
| Money markets and system funding | Seiche | Observed or unavailable |
| Capital markets | Seiche | Observed, derived, or unavailable |
| China economy | Palimpsest + Seiche | Revision-safe observations and bounded structural context |
| Covered bank and lender risk | LiquiLens | Observed, derived, or unavailable |
| GIFT City and gold | Seiche | Shared public context, source clocks and explicit partial or unavailable values |
| Forex | Seiche | Public reference context; no executable quotes |
| Market depth and exit liquidity | Undertow | Observed, derived, or unavailable |
Agent and terminal
Install the Agent Skill to let compatible agents activate this router for matching financial research. For direct terminal use, install the checksum-pinned Homebrew formula on macOS or Linux, or run the signed Git tag on any system with uv. Homebrew and the Git commands below install 0.1.7.
npx skills add https://github.com/beepboop2025/financial-evidence-skills/tree/v0.1.7/financial-evidence \
--skill financial-evidence
brew install beepboop2025/tap/financial-evidence
financial-evidence fetch --topic money-market --topic china-economy
uvx --from git+https://github.com/beepboop2025/financial-evidence-skills.git@v0.1.7 \
financial-evidence route --topic capital-market --format table
Market Brief is a free early-access companion for macro funding, capital-market, and liquidity research, with source dates and local comparisons. It does not score individual banks or lenders.
Shell-safe output is available as JSON, NDJSON, CSV, or a human-readable table. Exit code 0 means every retrieval succeeded, 1 means transport was partial, and 2 means every source was unavailable. Legacy status equals transport_status; both describe transport only, never evidence validity.
Use it where financial work already happens
Financial Evidence has native surfaces for Agent Skills, shell automation, remote MCP, Gemini CLI, Claude Code, a repository-installable Codex plugin, VS Code, Cursor, and MCP agents, plus compatibility layers for OpenBB, DuckDB, Excel Power Query, and FDC3 app directories. A ChatGPT public-directory submission is prepared but not submitted or listed. The compatibility files are open and self-hostable; they do not claim endorsement or marketplace listing by those vendors.
A public SKILL.md teaches compatible agents when to route money-market, capital-market, China-economy, bank-risk, and market-liquidity research across the four products.
Homebrew, uvx, persistent uv tools, completions for Bash, Zsh, and Fish, and stable pipeline exit codes.
Three read-only tools, an official registry record, multi-architecture container, root .mcp.json for Copilot Agent Host and compatible clients, and a one-click MCP Bundle.
Choose public research APIs and MCP, a private recurring research runtime, or an authenticated paper assessment host. Each interface preserves its own version, source clocks and access boundary.
Public Streamable HTTP at https://liquilens.in/mcp/financial-evidence, with no local runtime, account, API key, or write action.
A root extension manifest bundles the Agent Skill and allowlists the three tools from the public remote MCP endpoint.
A self-hosted liquidity-lab marketplace installs the versioned plugin, Agent Skill, and MCP configuration directly from the source repository.
Codex can install the repository plugin, which combines the same Agent Skill and public remote MCP route. ChatGPT availability remains pending account-owned universal-directory verification, submission, review, and publication.
A workspace servers config, protocol install URL, and code --add-mcp command, currently pinned to verified v0.1.7.
A workspace mcpServers config and base64-encoded install deeplink pinned to the same stdio server.
Registered routes for Python, notebook, REST, and MCP surfaces after installing the optional extension.
Six HTTP-backed views that keep the four products separate and queryable with SQL.
A Power Query function that returns topic evidence to ordinary workbooks without a custom add-in.
A public Evidence Inspector receives and broadcasts declared standard contexts through window.fdc3, with a matching FDC3 2.0 app-directory record.
https://liquilens.in/mcp/financial-evidence
gemini extensions install https://github.com/beepboop2025/financial-evidence-skills \
--ref v0.1.7
claude plugin marketplace add beepboop2025/financial-evidence-skills
claude plugin install financial-evidence@liquidity-lab
codex plugin marketplace add beepboop2025/financial-evidence-skills --ref v0.1.7
codex plugin add financial-evidence@liquidity-lab
code --add-mcp '{"name":"financial-evidence","type":"stdio","command":"uvx","args":["--from","git+https://github.com/beepboop2025/financial-evidence-skills.git@v0.1.7","financial-evidence-mcp"]}'
Install uv first, then review the exact local command and source before accepting editor install, trust, or tool-use prompts. Cursor asks for tool approval by default. VS Code normally asks you to trust a server on first start, but its inline mcp.json start action can skip that prompt. These are self-install routes, not vendor marketplace listings. The FDC3 record likewise does not imply FINOS review or directory acceptance.
Machine-readable: integration manifest · marketplace ledger · distribution status · llms.txt · pricing · MCP metadata
Evidence, not advice
It does not provide investment advice, security recommendations, execution quotes, credit ratings, or guarantees. It never converts missing, failed, restricted, or unavailable evidence into zero or calm. Transport success is not evidence evaluation or Evidence Carrier verification. Source publishers retain their rights, and fetched JSON is treated as untrusted data rather than executable instructions.
not_reported.Release verification record · Read the v0.1.5 semantic contract.