--- name: bohrium-mentor compatibility: "Requires bohr CLI (npm @dptech-corp/bohr-cli); verified with 2.9.19. Account operations require Bohrium authentication and network access." description: >- Ask Bohrium's AI science mentor with `bohr agents mentor` for scientific explanations, research-progress summaries, and method comparisons. Use for Bohrium Mentor, Science Navigator, AI 科研导师, 科学小导师, or literature-grounded scientific Q&A. Use `bohr paper`, `bohr scholar`, or `bohr lkm` for record-oriented retrieval and claim provenance. --- # Bohrium Mentor ## Choose the deliverable | Goal | Route | |---|---| | Explain a mechanism or compare methods | One focused Mentor question with scope and comparison criteria | | Obtain a screened corpus or evidence table | Paper retrieval/screening first; Mentor alone does not establish systematic coverage | | Check a specific claim or reference | Paper records or LKM provenance; verify the cited source | ## Workflow: produce and check a scientific explanation 1. Define the scientific question, domain, date scope and desired output. For a comparison, include the criteria (assumptions, accuracy, cost, applicability) in the question itself; these are question text, not new CLI flags. 2. Inspect the normalized input with `--dry-run` and the session price below. After the session budget is authorized, request structured output: ```bash bohr agents mentor "Compare DFT and Hartree-Fock for periodic solids: assumptions, accuracy, computational cost and limitations" --discipline Physics --journal-type foreign --model reason --yes -o json ``` 3. Keep the process attached for the answer. On successful JSON, retain `data.answer` and `data.session_id` together. The session ID is provenance, not an input to a CLI resume/get command. 4. Inspect the actual answer: an empty or visibly truncated `data.answer` is incomplete delivery even if the CLI returned `ok=true`; the CLI does not verify a service completion event. Check that it addresses every requested comparison criterion. Separate returned synthesis from independently checked evidence; resolve key references through paper/source retrieval when the deliverable requires verifiable citations. 5. Complete with the answer, uncertainty and actual verification boundary. If the stream fails or times out, report the partial/unknown outcome using the interruption rules below. Creating another session is a new billable action; do not treat retry as continuation of the original session. Use the version-matched `bohr agents mentor` command to ask one focused scientific question. Mentor is a synthesis workflow: it retrieves scientific literature internally and streams a Markdown explanation. Choose a more structured command when the deliverable is records rather than an explanation: - Use `bohr paper search` for specific papers and bibliographic fields. - Use `bohr scholar search` for scholar profiles. - Use `bohr lkm` for claims, reasoning chains, paper graphs, and provenance. - Use a dedicated literature-review workflow when the user needs a reproducible search protocol, screened corpus, or systematic evidence table. ## Authenticate and authorize the session Check the current identity and version-matched price before creating a session: ```bash bohr auth whoami -o json bohr billing pricing --resource mentor -o json ``` The bundled catalog price is **2.00 CNY per session**. The gateway precharges before session creation; answer-stream completion is not the charge trigger. A request to answer a question is not itself permission to spend. In non-interactive use, add `--yes` only after spending is authorized; otherwise return the `CONFIRMATION_REQUIRED` result to the user. Use `--dry-run` to inspect the normalized question and options without creating a session: ```bash bohr agents mentor \ "Compare density functional theory and Hartree-Fock for periodic solids" \ --discipline Physics \ --journal-type foreign \ --model reason \ --dry-run \ -o json ``` ## Ask the question After authorization, run one complete question: ```bash bohr agents mentor \ "Explain recent strategies for suppressing lithium dendrites in solid-state batteries" \ --discipline Chemistry \ --journal-type foreign \ --model reason \ --yes ``` `All`, `Physics`, `Chemistry`, `Biology`, `Materials`, and `Computer Science` are common discipline values. The CLI forwards filter values to Mentor; keep the default `All` when the requested field is uncertain. `foreign` and `chinese` select the journal scope. `reason` is the default model. For automation, prefer JSON: ```bash bohr agents mentor "Explain the evidence for high-temperature superconductivity" \ --yes \ -o json ``` Human output streams only the main Markdown answer. Structured success exposes `data.answer` and `data.session_id`. Treat inline citation markers as part of the answer, not as complete bibliographic records; use `bohr paper search` when the user needs independently inspectable references. ## Handle delay and interruption Deep reasoning may have a noticeable first-token delay, and the command timeout is five minutes. Keep the process attached while useful output is still arriving. Session creation and answer streaming are separate phases. If streaming fails, the session may already be billable and human output may already contain a partial answer. Preserve the actual stdout, structured error, billing notice, and request ID that were returned. A successful machine result is the only place the CLI returns `data.session_id`. **Do not automatically retry** after session creation or a stream failure. A retry creates a new session and may charge again. Report the interruption and ask whether the user authorizes another session. ## Capability boundary The CLI intentionally renders only the main answer. It does not expose Mentor's non-main UI layouts, citation-card metadata, recommended follow-ups, or a resume command. Stay inside `bohr agents mentor`; do not reconstruct the service's transport protocol or credential handling in shell or Python. The old top-level `bohr mentor` path has been removed. Use `bohr agents mentor`. ## Choose the payment source Without `--payment-source`, the CLI sends no channel preference: the gateway uses photons first and falls back to balance only on definite photon insufficiency. Tell the user this policy when obtaining spending authorization. To restrict spending to one source, use `--payment-source photon` or `--payment-source balance`. Explicit photon stops when photons are insufficient; explicit balance charges balance directly. Existing free quotas and exemptions still apply. When an explicitly selected channel returns `BILLING_INSUFFICIENT` with `retryable: false`, report the refusal and wait for the user to recharge the selected source or authorize the other source for the original command. Explicit balance insufficiency also stops without charging photons; use `--payment-source photon` only after that source is authorized. `--yes` confirms execution; it does not authorize changing the payment source. Use `--payment-source balance` only when that source is authorized. Keep the selected source on subsequent billable calls in a multi-step workflow; options from a previous CLI invocation are not saved. `bohr api` follows the same rule. ## Detailed reference Read [`references/commands.md`](references/commands.md) for exact flags, output, billing, dry-run behavior, failure handling, and the embedded-version contract.