Mira workflow
A working repository for Claude Code and Codex where an investment thesis is backed by sources, has disconfirming conditions and a refresh date
- 280 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The agent visits public sources and APIs and writes research packages into the project folder
- For a position or portfolio review you hand the agent holdings, weights and constraints, which is personal financial data
- Outputs can be incomplete or stale, and decisions based on them are yours
Install
- Works in
Manual install
git clone https://github.com/byteseek/mira.gitOpen the cloned folder as a project in Claude Code or Codex and write "hi Mira". If the product does not load the repository rules, paste the text from `docs/chatgpt-conversation-instructions.md`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent visits public sources and APIs and writes research packages into the project folder
- For a position or portfolio review you hand the agent holdings, weights and constraints, which is personal financial data
- Outputs can be incomplete or stale, and decisions based on them are yours
This is third-party code. Review the repository files before installing.
What it does
Mira is a set of rules, loops and templates that you open as a project in Claude Code, Codex or another code agent. The agent enters Mira mode on "hi Mira" or on a concrete research task and then works under the contract in `OPERATING_CONTRACT.md`. The model widens attention here: it reads, compares, extracts claims and drafts. The purpose, framing, loss tolerance, final judgment and any market action stay with the human. Every conclusion comes with an evidence log: what is a fact, a claim, an assumption, an opinion and a calculation. For a thesis it records what supports it, when it is valid, what would disconfirm it and when the conclusion is considered stale. The router `loops/analysis-routing.md` sends a task to the right loop: first-pass equity research, monitoring of an existing thesis, a daily or weekly market brief, an earnings report, interpretation of analyst reports, an industry, macro, ETFs, or a position or portfolio review. The `cases/` folder holds ready packages as examples: AAPL research, the NVTS quarter, the ABF industry and ETF discovery. `scripts/check_updates.sh` checks the repository's freshness and `scripts/mira_update.sh` updates it.
Who it is for. For private investors and analysts who want to run research with an agent while keeping sources, assumptions and review conditions.
Good fit when
- You need first-pass research on a stock or industry delivered as a memo, evidence log and expectation map
- A company reported earnings or guidance and you must work out what changed in the thesis
- You need to interpret an analyst report: extract claims, valuation assumptions and gaps against expectations
- You want a daily or weekly market brief that separates fact, drivers of moves and the event calendar
- An existing thesis must be refreshed regularly, separating new data from data that changes the conclusion
Not a fit when
- You need trade signals or automated trading: Mira is not a trading bot or a portfolio manager
- You need a personal recommendation on what to buy: the repository leaves the decision to the human and gives no investment advice
- You need a real-time market data feed: the project has no data daemon
- You need a one-line quick answer: a formal package requires an evidence log and a refresh date
Example request
hi Mira. Build first-pass research on Coherent: thesis, evidence log, what would disconfirm it and when to refresh itLimitations
The repository is meant for research process design, documentation and historical examples. It gives no investment, legal, tax or accounting advice. The same applies to any agent output produced with it: treat it as support material, not a recommendation and not a verified fact.
Public case packages and generated materials may be incomplete, inaccurate or stale after the date stated in them. Each conclusion gets a refresh boundary, and past it the conclusion must not be reused without checking.
Data sources, APIs and third-party exports come in through a controlled ingestion layer that checks license, provenance and calculations. For a position or portfolio review you supply the holdings, weights and constraints yourself. The main README is in English, with a Chinese version.
How to disable. Close the project or delete the cloned `mira` folder. The agent stops entering Mira mode. The packages and logs you built live inside the folder, so save them first.
FAQ
How do I start working with Mira in Claude Code or Codex?
Clone the repository and open it in the agent. Write "hi Mira" or give a research task right away. For the greeting the agent returns a short summary of `START_HERE.md`, and a concrete task is routed straight through the router.
What is an evidence log?
A claim-level table: source, record type (fact, claim, assumption, opinion, market pricing or calculation) and review conditions. The format is described in `data/evidence-log-schema.md`.
What is a refresh boundary?
The condition after which a conclusion is unsafe to reuse: a `stale_after` date or events in `must_refresh_if`. Time rules are collected in `data/time-policy.md`.
Does Mira trade for me or give recommendations?
No. The project calls itself a research system, not a trading bot. Decisions, loss limits and any market action stay with the human.
How do I update Mira?
Run `scripts/mira_update.sh`. The script fetches changes, refuses to work with a dirty, ahead or diverged tree, applies only fast-forward and validates the repository. The freshness check `scripts/check_updates.sh` updates nothing.
Discussion
No questions or reviews yet.
Related
WorkflowsA Claude Code plugin for equity research reports: fundamental and technical analysis, scenarios and risk assessment, for education only
16 Claude Code commands for Russian sole proprietors: invoices, acts, contracts, taxes and contributions, with data kept in local files
A set of skills for individual investors and traders: market review, stock screening, technical analysis, trade planning, journaling and post-trade review
A set of investment research skills for Claude Code and Codex built on four investors' methodologies. It is a research tool, not investment advice
Anthropic's plugin marketplace for finance teams: 10 agents plus skills and MCP connectors for DCF and LBO models, earnings, reconciliation and KYC
Claude for Financial Services CN
claude-for-financial-services-cn
An unofficial set of 63 skills for Claude Code: DCF, comps, LBO, notes and decks for Chinese equities with Wind, iFind and AkShare data
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| Mira (this entry) | no | 280 | |
| Equity Research Skill | no | 468 | a skill that writes a single nine-chapter stock report with a reverse DCF rather than maintaining a thesis with an evidence log and refresh date |
| Claude Equity Research | no | 749 | a plugin for stock analysis reports, without a persistent thesis or review conditions |
README in short
The document lays out a division of labor between human, model and system, then links to the agent entry cards and a quick start. It continues with tables of task routes, concepts, a documentation map and examples.