AGENT SKILL

sell-research-honestly

Audit, position, and communicate research value without outrunning the evidence. Use when users need audience-specific buy-in from advisors, collaborators, reviewers, talk audiences, or the research community; want to pitch or promote an idea, paper, result, codebase, or demo; plan research PR; or need to detect and repair overclaiming. Do not use for generic proofreading, summarization, or formatting.

License 许可
MIT
最近更新
2026-11-14
Stars
★ 3
Forks
0

01 前言 Front Matter

name
sell-research-honestly
description
Audit, position, and communicate research value without outrunning the evidence. Use when users need audience-specific buy-in from advisors, collaborators, reviewers, talk audiences, or the research community; want to pitch or promote an idea, paper, result, codebase, or demo; plan research PR; or need to detect and repair overclaiming. Do not use for generic proofreading, summarization, or formatting.

02 资源文件 2

03 SKILL.md 正文

Sell Research Honestly

Turn one research object into the strongest defensible case for one audience and one desired action. Keep facts invariant while changing emphasis.

Inputs

Establish three inputs:

  1. Research object — an idea, experiment, paper, codebase, talk, demo, or the files that contain it.
  2. Audience — one advisor, collaborator, reviewer, talk audience, or research community.
  3. Desired action — what that audience should understand, approve, discuss, accept, use, cite, or do next.

Inspect supplied materials and repository evidence before requesting facts. If an input remains missing, ask one blocking question per turn while completing the audit work that is already possible.

1. Frame the case

Classify the research object as idea, preliminary, or evidenced. Read the matching audience section in references/audience-playbooks.md, then write a four-part audience model:

  • outcome the audience values;
  • cost or decision risk they bear;
  • objection they are most likely to raise;
  • evidence that would make the desired action reasonable.

Read the reference's feedback section only when the user brings back real reactions or outcomes.

Complete when: the research object, maturity, single audience, desired action, and all four audience-model fields are explicit.

2. Build the value case

Create a six-field value case. Mark every field supported, plausible, missing, or contradicted.

Field Question
Problem What concrete problem exists?
Audience stake Why does this audience care?
Gap Where do current approaches fail or cost too much?
Insight / solution What new information or intervention does the work add?
Evidence What makes each material claim credible?
Boundary / cost Where does it fail, stop applying, or impose a cost?

Build a claim ledger beside the value case. Give every material claim one state and an exact evidence pointer:

  • observed — directly supported by the materials;
  • inferred — a reasoned interpretation of observed evidence;
  • hypothesized — a proposition awaiting a test;
  • unsupported — no supplied support;
  • contradicted — conflicts with supplied evidence.

Prefer evidence in this order: raw results, data, code output, figures, and verified citations; locatable project or manuscript material; user-provided but unverified information; agent inference. Label the last two categories. Treat novelty, priority, state-of-the-art, and comparisons with current work as external facts: verify them against external literature. Until verified, keep their claim state unsupported, add verification: unverified, and exclude them from public copy.

Make the evidence burden proportional to the claim:

  • Attach evidence to strong terms such as first, general, robust, low-cost, and state of the art; otherwise use a supported term.
  • Inspect critical subgroup and failure behavior rather than relying on an average alone.
  • Account for added compute, data, labor, latency, memory, and deployment cost.
  • Match causal language to a design that excludes the main alternative explanations.
  • Match scope to the tested models, datasets, conditions, and cases.
  • Require fair baselines for comparative effectiveness.

When evidence is narrower, narrow the claim.

Complete when: all six fields have a status, every material claim has one state and a source pointer, and every comparative external claim is verified or held unsupported with verification: unverified.

3. Apply the value gate

Choose exactly one verdict:

  • ready-to-sell — the value proposition and material claims are supported;
  • narrow-and-sell — a useful case survives after narrowing or qualification;
  • validate-first — the value is plausible but a load-bearing claim lacks evidence;
  • do-not-claim — the core claim is unsupported, contradicted, or relies on a research-integrity breach.

For validate-first, produce a minimum discriminating validation plan: claim, cheapest informative test, success and failure criteria, and how each outcome changes the message. Treat the result as an internal validation brief rather than public-ready copy.

For do-not-claim, run one salvage pass for a narrower phenomenon, negative result, failure mode, engineering benefit, cost advantage, or worthwhile research question. Use the recovered case only if its own claim ledger passes the gate. If nothing survives, make the Primary Deliverable a concise no-case finding that names the evidence required to reopen the claim.

The verdict measures readiness for honest communication, not the probability of acceptance, collaboration, publication, or attention.

Complete when: the verdict follows from the value case and fixes which deliverable is allowed next.

4. Compose the strongest defensible message

For ready-to-sell or narrow-and-sell, use this narrative spine:

audience-relevant problem → concrete gap → insight or solution → value → strongest evidence → boundary and cost → explicit next action

Write observed claims as facts. Mark inferred claims with calibrated language. Present hypothesized claims only as research questions or future tests. Replace unsupported and contradicted claims with the narrowest supported alternative.

Vary emphasis, length, and vocabulary for the audience; keep evidence, costs, and scope unchanged. Produce one primary deliverable unless the user requests additional formats. Name the concrete contribution before any larger vision.

Complete when: the deliverable serves the audience model, contains one clear takeaway and next action, and every factual sentence maps to the claim ledger.

5. Run the integrity pass and deliver

Re-run the Step 2 evidence rules against every sentence in the draft. Repair each mismatch by adding evidence, qualifying the sentence, or replacing it with the narrowest supported claim.

Return four sections:

  1. Verdict — status and decisive reason.
  2. Value Case — the six audited fields and audience model.
  3. Primary Deliverable — one ready-to-use artifact allowed by the gate.
  4. Integrity Notes — claim ledger, required qualifiers, evidence gaps, and next action.

Default to chat output. Edit only files the user explicitly names. Draft PR plans, outreach, and posts locally; publish, send, or contact people only after the user authorizes that external action. Match the user's language unless the target artifact establishes another language.

Complete when: the audience and action are explicit; the verdict matches the research maturity; every material claim is traceable; the deliverable preserves boundaries and costs; and the next action is concrete.

04 同仓库更多 Skills

pengqianhan
pengqianhan/AI-Human-Research-OS

workflow-skill-creator

Distills a completed user workflow or interaction into a reusable agent skill. Use when the user asks to turn their workflow, interaction, or multi-step process into a skill, or when they say "make this a skill", "create a skill from what we just did", "package this workflow" or similar. Do not use for creating skills from scratch without an existing workflow (use a generic skill-creator for that).

★ 3
pengqianhan
pengqianhan/AI-Human-Research-OS

uv

Checks whether the uv Python package manager is installed and installs it if missing. Ensures uv is on PATH. Use when another skill requires uv as a prerequisite.

★ 3
pengqianhan
pengqianhan/AI-Human-Research-OS

uniprot-database

Access protein metadata, function, taxonomy, and sequences across UniProtKB, UniParc, and UniRef. Use when searching for proteins, mapping identifiers, or retrieving functional annotations and publications. Don't use for sequence alignment, protein folding, or sequence similarity search (use specialized skills for those tasks).

★ 3
pengqianhan
pengqianhan/AI-Human-Research-OS

unibind-database

Queries the UniBind database for experimentally validated transcription factor (TF) binding sites. Use when retrieving direct TF-DNA interaction datasets, downloading binding site coordinates (BED/FASTA) for local analysis, or listing available datasets by species, cell line, or TF name. Don't use to query specific intervals, locations, genes, motif models or expression data.

★ 3
pengqianhan
pengqianhan/AI-Human-Research-OS

ucsc-conservation-and-tfbs

Fetch Evolutionary Conservation scores (phyloP, phastCons) and Transcription Factor Binding Sites (TFBS) from the UCSC Genome Browser. Use when analyzing whether genomic variants or regions are evolutionarily conserved, functionally important, or bounded by TF regulators across major projects (ENCODE, JASPAR, ReMap).

★ 3
pengqianhan
pengqianhan/AI-Human-Research-OS

string-database

Query the STRING database for protein-protein interactions (PPIs), functional enrichment, and homology. Use when the user asks about interactions between specific proteins, interaction evidence, confidence scores, protein interaction partners, or pathway enrichments.

★ 3