llms.txt@web/public git:20261015.8f8cbf3A

Immutable. This exact content never changes and is served at /api/v1/blob/65cf9609a21724ab.

# Bounty Operator

> Pre-submission review for bug bounty reports and smart-contract findings. The AI argues against the finding. The hunter writes the report. Reviews run on the user's own AI provider API key. Free: 1 review a day. Operator: US$10 per month for unlimited reviews and US$2 API credit per paid month. Only hosted reviews count: on the site or through `run_review`.

Built by Tradi3, 77th on Immunefi's all-time leaderboard on 10 October 2026 (profile: https://immunefi.com/profile/Tradi3/): 2nd of 133 in Immunefi's Firelight audit competition, 8th of 65 in Quantus, and 15th of 186 in ENS with 17 valid Critical submissions, the most on that board (leaderboards: https://immunefi.com/audit-competition/audit-comp-firelight-1/leaderboard/, https://immunefi.com/audit-competition/audit-comp-quantus/leaderboard/, https://immunefi.com/audit-competition/audit-competition-ens/leaderboard/). Platform-neutral: Immunefi, Cantina, Sherlock, HackerOne. Every URL below is canonical and extensionless. Source code: https://github.com/bountyoperator/bounty-operator (MIT). The three core profiles, the free tools, the CLI and the MCP server are in that repository. The gauntlet stages and the panel cross-examination run on the hosted service.

A hosted review sends the user's files and provider key through bountyoperator.com to the provider the user chose. The server adds the review method and stores no files, prompts, keys or results. For a hosted profile that method goes only to the provider the user chose, which with OpenRouter means OpenRouter and the model host it passes the call to: no page, download or tool returns it.

## Product

- [Model API](https://bountyoperator.com/model-api): Hosted cybersecurity and low-refusal model API (`/v1/chat/completions`, `/v1/models`) with prepaid credit, per-key spending limits, function tools, streaming, workload cost estimator and interactive quickstart, plus native Infron 7-model cybersecurity stack support (`provider: "infron"`, `INFRON_API_KEY`) in the workbench, Gauntlet, Panel review, CLI and MCP. No subscription required; the monthly Operator plan includes US$2 API credit per paid month. Optional auto top-ups are off by default.

- [Benchmark](https://bountyoperator.com/benchmark): Paydirt, the model benchmark: which model finds the planted bug, leaves the fixed twin alone and catches an overclaimed report, with cost per run.
- [Benchmark method](https://bountyoperator.com/benchmark/method): How Paydirt is built and scored, and how to check every published number.
- [Workbench](https://bountyoperator.com/#workspace): Select files, pick a review profile and run it on your own AI provider API key. The three core profiles also export as a prompt.
- [Pricing](https://bountyoperator.com/pricing): Free gives you 1 review a day, any review type (reset at 00:00 UTC). Operator is US$10 per month for unlimited reviews, the Gauntlet, Panel review and 4 reviews at once. Includes US$2 in Bounty Operator API credit per paid month; other providers bill usage separately. Existing weekly subscriptions keep their Stripe billing interval and do not receive the monthly credit. The review of the day is a hosted review, on the site or through `run_review`. Copy-paste reviews in a chat app, and core reviews the agent's own model writes through `prepare_review` or the skills, need no account and are not counted.
- [Method](https://bountyoperator.com/method): Why bug bounty reports get closed. Twelve checks, the question each one asks, why reports die on it and the verdict it leads to.
- [Gauntlet](https://bountyoperator.com/gauntlet): One run through eight stages (scope, provenance, prior art, proof, severity, triager, report, verdict) that ends in one verdict.
- [Panel review](https://bountyoperator.com/panel-review): Two to four models review in parallel, then a cross-examination pass keeps what the cited lines prove.
- [Your model, your key](https://bountyoperator.com/your-model-your-key): Where your code goes, who holds the API key and what is stored.
- [Compare](https://bountyoperator.com/compare): Sourced tables against chat apps, open-source audit skills, platform pre-checks and seven overlapping tools.

## Review profiles

- [Challenge a draft report](https://bountyoperator.com/challenge-report): Every claim in a draft checked against the code it cites. Core profile, id `report`.
- [Solidity review](https://bountyoperator.com/solidity-review): Entry points, invariants and value flow in Solidity contracts, with a line reference for every finding. Core profile, id `solidity`.
- [Code security review](https://bountyoperator.com/code-security-review): Authorization, injection and failure paths in any codebase, each finding with file, line and a fix. Core profile, id `general`.
- [Triager simulation](https://bountyoperator.com/triager-simulation): The likeliest reasons a triager closes the report, and the evidence that answers each. Hosted, id `triage`.
- [Prior-art overlap](https://bountyoperator.com/prior-art-check): A finding compared with the known issues and audits you supply: same root cause, same symptom or unrelated. Hosted, id `prior-art`.
- Six more hosted profiles run from the workbench or through `run_review`: scope and impact fit (`scope`), design intent and actors (`provenance`), proof review (`poc`), severity calibration (`severity`), report editor (`report-edit`) and scanner triage (`scanner`).
- Core: the method is open source, and the profile runs hosted, as an exported prompt or over MCP on the agent's own model. Hosted: the profile runs on bountyoperator.com with the user's provider key and has no prompt export.

## Free tools

- [All tools](https://bountyoperator.com/tools): Free tools that run in the browser. Your content never leaves it.
- [Invalid report costs](https://bountyoperator.com/invalid-report-costs): What a report closed as invalid, N/A or spam costs on HackerOne, Bugcrowd, Intigriti, YesWeHack, Immunefi, Cantina and Sherlock, from each platform's own rules, with dated sources.
- [Report check](https://bountyoperator.com/tools/report-check): Instant checks on a pasted draft report, in the browser, including the 2026 platform rules: no transaction sent to a live network, AI use disclosed, no placeholders or assistant leftovers.
- [Packet verifier](https://bountyoperator.com/tools/verify): Recompute the SHA-256 of every file in a review packet.
- [Secret check](https://bountyoperator.com/tools/secret-check): Find keys, tokens and seed phrases in a PoC or gist before publishing it.
- [Acceptance rates](https://bountyoperator.com/tools/acceptance-rates): Acceptance rates by smart-contract bug class from public contest judging.
- [Slither triage queue](https://bountyoperator.com/tools/slither-focus): Slither JSON reduced to a ranked queue of detectors worth a manual pass.

## Report templates

- [Report templates](https://bountyoperator.com/templates): Templates for four platforms and a Foundry PoC scaffold.
- [Immunefi](https://bountyoperator.com/templates/immunefi): Immunefi bug report template in Markdown.
- [Sherlock](https://bountyoperator.com/templates/sherlock): Sherlock audit contest report template in Markdown.
- [Cantina](https://bountyoperator.com/templates/cantina): Cantina finding report template in Markdown.
- [HackerOne](https://bountyoperator.com/templates/hackerone): HackerOne bug report template in Markdown.
- [Foundry PoC](https://bountyoperator.com/templates/foundry-poc): Foundry fork-test scaffold with a control run and a final assertion.

## Guides

- [Model API setup guide](https://bountyoperator.com/docs/model-api): Connect agents and chat tools, discover models, send requests and understand prepaid billing.
- [Choose a model](https://bountyoperator.com/docs/model-selection): Compare workflow needs, capabilities, cost, availability, data handling and evaluation evidence.

- [Report guide](https://bountyoperator.com/guide): How to write a bug bounty report that survives triage, with a worked example rewritten fragment by fragment.
- [MCP server](https://bountyoperator.com/mcp): Install commands for Claude Code, Codex and Cursor, the tools, the three prompts and token setup.
- [Changelog](https://bountyoperator.com/changelog): Every release with its date.

## Trust and legal

- [Security](https://bountyoperator.com/security): Data flow, what is stored, the headers and CSP in force, and how to report a vulnerability.
- [Privacy policy](https://bountyoperator.com/privacy): What is stored, what is never stored, the two cookies, and account export and deletion.
- [Terms of service](https://bountyoperator.com/terms): Plans, monthly renewal, legacy weekly subscriptions, API credit, optional auto top-ups, cancellation, refunds and the rules of use.
- [Licences](https://bountyoperator.com/licenses): The MIT licence and the open-source packages the Worker depends on. The MCP server has no dependencies.

## For agents

- MCP endpoint (Streamable HTTP, JSON-RPC over POST): https://bountyoperator.com/api/mcp
- Add it to Claude Code: `claude mcp add --transport http bounty-operator https://bountyoperator.com/api/mcp`
- Add it to Codex: `codex mcp add bounty-operator --url https://bountyoperator.com/api/mcp`
- The remote endpoint has five tools. With no account: `list_profiles`, `prepare_review`, `build_packet`. With an account token: `account`, `run_review`. Three prompts: `challenge-report`, `solidity-review`, `gauntlet`.
- `list_profiles` returns every profile with `hosted`: false for a core profile, true for a hosted one.
- `prepare_review` takes the three core profiles (`general`, `solidity`, `report`) and returns the method and the review request for the agent's own model. For a hosted profile it fails with the code `hosted_profile` before any file is read.
- `run_review` runs any profile as a hosted review and is the only way to run a hosted one. It needs the account token in the `Authorization` header and the provider key in the `X-Provider-Key` header, and uses one hosted review. A review the provider blocks is not counted: the result carries `blocked` (`anthropic-cyber`, `anthropic-reasoning`, `openai-cyber`, `guardrail` or `policy`) with `refused`, or the call fails with the code `provider_policy`. The text is then not a review. Do not call the same model again; use another model or provider (https://bountyoperator.com/guide#model-refuses).
- The `gauntlet` prompt runs the eight stages in order: seven hosted through `run_review`, and the report stage through `prepare_review` on the agent's own model. It needs an account token. Seven hosted reviews are more than the free review of the day, so a full run takes Operator.
- A failed tool call returns `isError` with `{ error, code }`. Codes: `hosted_profile`, `privacy_block`, `privacy_warn`, `daily_used`, `operator_only`, `review_running`, `provider`, `provider_policy`, `output_withheld`. `operator_only`: `run_review` was called with `verdict` or `panel` on an account without Operator; nothing was sent and the review of the day is not used. `output_withheld`: the model repeated its instructions instead of reviewing, so the answer was stopped and the call used one review; run it again or choose a stronger model.
- Limits per call: 50 files, 120 KB per file, 240 KB and 20,000 lines in total.
- A hosted review runs for up to 15 minutes. claude.ai and Claude Desktop stop any tool call after 240 seconds, so run hosted reviews from Claude Code (server entry `"timeout": 1200000`), Codex (`tool_timeout_sec = 1200`) or the website. `prepare_review` and `build_packet` for the core profiles work in every client.
- Local MCP server: `npx -y https://bountyoperator.com/dl/bounty-operator-mcp.tgz`. It has six tools: it adds `run_gauntlet_plan` (no account) and reads files by path. SHA-256 of the package: https://bountyoperator.com/dl/SHA256SUMS.txt
- Skills and Claude Code plugin: `claude plugin marketplace add bountyoperator/bounty-operator` then `claude plugin install bounty-operator@bounty-operator`; any agent: `npx skills add bountyoperator/bounty-operator`. Five skills: challenge-report, solidity-review and code-security-review run the core methods on the agent's own model; gauntlet runs the hosted stages over MCP; hunt-with-gate is manual only and never submits.
- Immunefi Studio's own MCP server, https://studio.immunefi.com/api/mcp, takes a personal access token from https://studio.immunefi.com/agents. When an agent has it connected, the gauntlet and challenge-report prompts and skills find the Immunefi programme with its `list_programs` tool and, for a contract, call `get_proxy_history` and `get_target_state`, and pass what they return to the review as files. Setup: https://bountyoperator.com/mcp#immunefi
- Health: https://bountyoperator.com/api/health returns `status`, `version`, `reviews` and `profiles` (`hosted` on this site).
- Sitemap: https://bountyoperator.com/sitemap.xml
- Security contact: https://bountyoperator.com/.well-known/security.txt

## Contact

- [Help and feedback](https://bountyoperator.com/support): Where each kind of request goes. Questions and ideas go to the public forum on GitHub Discussions, bugs to the issue form, account and payment questions to email.
- Forum: https://github.com/bountyoperator/bounty-operator/discussions (public: no unreported findings, report drafts or keys)
- Support: support@bountyoperator.com
- Security: security@bountyoperator.com