agentleFS
eunomia-bpfeunomia-bpf/eunomia.dev/.agents/skills/eunomia-research-report/SKILL.md

eunomia-research-report

Skill244 starsChanged 5 days agoMITView on GitHub

Research, write, validate, and publish source-grounded Eunomia Daily Reports for technical readers. Use when an agent needs to use platform Deep Research or direct primary-source research, identify a non-trivial systems gap, propose high-quality academic and production ideas, write bilingual reports under the stable `/research/` URLs, or revise an existing Daily Report.

The file

Show its rules and text

We split this skill into the rules it gives an agent

34 rules. About 0 are common, 32 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈32

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every Daily Report must satisfy these rules:

    line 14
  2. Distinctive:

    Every scheduled daily run publishes one new report.

    line 21
  3. Distinctive:

    Do not add provenance banners, warning boxes, generation-process badges, review-status text, or footer disclaimers to the public page.

    line 23
  4. Distinctive:

    Do not infer an article byline from Git commit metadata.

    line 24
  5. Distinctive:

    Read these before acting:

    line 28
  6. Distinctive:

    When invoked by a scheduled repository task, also read the task entrypoint, current operating records, rolling topic mix, and any explicit …

    line 38
  7. Distinctive:

    Keep an existing URL when revising an article.

    line 47
  8. Very specific:

    prefer the active series in .github/seo-data/content-series.md.

    line 70
  9. Distinctive:

    keep 5–7 of the most recent 10 reports centered on eBPF.

    line 71
  10. Distinctive:

    keep pure Agent reports to at most 1–2 of 10.

    line 72
  11. Distinctive:

    use the remaining slots for adjacent systems topics such as Linux, profiling, networking, security, runtimes, GPU/heterogeneous systems, …

    line 73
  12. Distinctive:

    Do not choose a topic merely because it is trending or can mention an Eunomia project.

    line 94
  13. Distinctive:

    When platform Deep Research is available or requested, use it for broad source discovery, competing explanations, and citation collection.

    line 98
  14. Distinctive:

    Treat its report as research input, not publishable prose.

    line 100
  15. Distinctive:

    verify dates, metrics, experimental conditions, and quoted conclusions.

    line 103
  16. Distinctive:

    Before drafting, state in ordinary language:

    line 127
  17. Distinctive:

    If the candidate fails this gate, discard it and immediately research the next candidate in the approved series roadmap (see Daily …

    line 139
  18. Distinctive:

    Use a concrete recurring scenario before introducing a taxonomy, formal model, or coined term.

    line 143
  19. Distinctive:

    Do not front-load a source matrix, equation, related-work parade, or newly named property.

    line 157
  20. Distinctive:

    Use a reader-facing heading such as:

    line 161
  21. Distinctive:

    Do not repeat individual papers' future-work paragraphs.

    line 173
  22. Distinctive:

    Use a heading such as:

    line 177
  23. Distinctive:

    Prefer two or three developed ideas over a brainstorm list.

    line 182
  24. Distinctive:

    use Chinese for ordinary concepts.

    line 201
  25. Distinctive:

    do not make the grammar depend on a stack of English nouns.

    line 203
  26. Very specific:

    Do not publish headings such as Scope, limitations, and falsification or direct Chinese translations of that template.

    line 208
  27. Distinctive:

    Use a natural heading such as:

    line 210
  28. Distinctive:

    Every scheduled daily run must finish with one new bilingual report deployed publicly.

    line 219
  29. Distinctive:

    create a focused branch;

    line 221
  30. Distinctive:

    add exactly one new Daily Report topic in English and Chinese.

    line 222
  31. Distinctive:

    run the repository's full verification workflow.

    line 224
  32. Distinctive:

    verify that the exact merge commit deploys.

    line 227
  33. Distinctive:

    If the initial candidate lacks a defensible gap, mechanism, evidence base, or non-trivial idea:

    line 254
  34. Distinctive:

    if necessary, move to another approved eBPF or adjacent-systems series while respecting the rolling mix.

    line 259

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Eunomia Daily Report

Produce technically serious, source-grounded analysis for the public Daily Report section. The repository is the source of truth for the publication contract, content, routes, topic mix, and delivery workflow.

The purpose is to help engineers and researchers understand a real problem, see what current work still misses, and find ideas worth implementing and evaluating. Depth comes from evidence, mechanism, comparison, and testable design, not from academic tone or terminology density.

Public Contract

Every Daily Report must satisfy these rules:

  • The title and opening are understandable to a technically qualified reader who has not read the source corpus.
  • Facts, measurements, cited claims, synthesis, and proposed ideas remain distinguishable.
  • The report contains a concrete section explaining where current research or production practice is still weak.
  • The report contains a small number of developed directions with both academic and production relevance.
  • The report states the assumptions behind its conclusion and the evidence that would change it, using reader-facing language.
  • Every scheduled daily run publishes one new report. A weak candidate is rejected, but the run must continue researching another approved question until one passes the quality gates.
  • The rolling topic mix in .github/seo-data/content-series.md is mandatory: normally 5–7 of the most recent 10 reports explicitly center eBPF, while pure Agent topics remain at most 1–2 of 10.
  • Do not add provenance banners, warning boxes, generation-process badges, review-status text, or footer disclaimers to the public page.
  • Do not infer an article byline from Git commit metadata.

Required Context

Read these before acting:

  • CLAUDE.md
  • .agents/skills/blog-writing-style/SKILL.md
  • .github/seo-data/content-series.md
  • existing pages under docs/research/
  • related posts under docs/blog/ and docs/reports/
  • references/research-method.md
  • docs/papers/registry.yaml only after a candidate question exists

When invoked by a scheduled repository task, also read the task entrypoint, current operating records, rolling topic mix, and any explicit publication constraints stored in the repository.

Output Location And Metadata

A bilingual report uses:

  • docs/research/<topic>.md
  • docs/research/<topic>.zh.md

Keep an existing URL when revising an article. Scheduled daily publication, however, requires one new report page; revising an existing page does not satisfy that day's publication requirement.

New Daily Report frontmatter should include:

date: YYYY-MM-DD
title: "Reader-facing title"
description: "Concrete problem, main analysis, and practical consequence."
tags:
  - Daily Report
  - <precise topic tags>
research_question: "The exact question the report tries to answer"
source_cutoff: YYYY-MM-DD
status: daily-report

Topic Selection

Daily Report is eBPF-first.

Before research begins:

  1. calculate the rolling mix from the most recent published reports;
  2. prefer the active series in .github/seo-data/content-series.md;
  3. keep 5–7 of the most recent 10 reports centered on eBPF;
  4. keep pure Agent reports to at most 1–2 of 10;
  5. use the remaining slots for adjacent systems topics such as Linux, profiling, networking, security, runtimes, GPU/heterogeneous systems, distributed systems, storage, and compilers.

An eBPF-centered report must make eBPF essential to the mechanism, comparison, runtime boundary, or experiment. Mentioning eBPF in the introduction or future-work section does not make a report eBPF-centered.

Because the current small archive already contains two Agent-centered reports, the next reports should strongly favor eBPF before another pure Agent report is considered.

Workflow

1. Start With A Real Reader Question

Begin from the active eBPF or approved adjacent-systems series, then search before choosing a thesis. Prefer a question that changes an architecture, implementation, security, operations, or research decision.

Good questions expose one of these:

  • a mechanism missing from current systems;
  • conflicting results that need reconciliation;
  • a production failure absent from benchmarks;
  • an abstraction that breaks under new workloads;
  • a capability that exists but lacks a correct interface;
  • a measurement problem that prevents comparing alternatives.

Do not choose a topic merely because it is trending or can mention an Eunomia project.

2. Use Platform Deep Research As Input

When platform Deep Research is available or requested, use it for broad source discovery, competing explanations, and citation collection.

Treat its report as research input, not publishable prose. Before using it:

  • inspect the important primary sources;
  • verify dates, metrics, experimental conditions, and quoted conclusions;
  • remove unsupported extrapolation;
  • identify which sources are independent;
  • compare the candidate with existing Eunomia content;
  • rebuild the argument around the reader's question rather than the tool's outline;
  • record internally when the platform report was incomplete or unavailable.

3. Build An Evidence Map

For every serious source, capture privately:

  • source type and primary URL;
  • publication date and event date;
  • concrete fact, mechanism, or measured result;
  • workload, scale, method, and assumptions;
  • implementation or artifact availability;
  • possible conflict of interest;
  • independent support or counterevidence;
  • which reader decision it changes.

Count independent evidence, not repeated coverage. A focused report should use the smallest corpus that adequately supports the mechanism, alternatives, gap, and boundary. Do not pad a narrow report to imitate a survey.

4. Pass The Thesis And Gap Gate

Before drafting, state in ordinary language:

  1. the concrete problem;
  2. the current default mental model;
  3. why that model fails;
  4. the central claim;
  5. the strongest alternative explanation;
  6. the research or production gap exposed by the evidence;
  7. what result would change the claim.

The gap must identify a missing benchmark, interface, mechanism, guarantee, dataset, measurement, deployment property, or boundary condition. Reject generic statements such as "more research is needed" or "scalability remains challenging."

If the candidate fails this gate, discard it and immediately research the next candidate in the approved series roadmap (see Daily Fallback).

5. Design The Reader Path

Use a concrete recurring scenario before introducing a taxonomy, formal model, or coined term.

A common progression is:

  • concrete situation and consequence;
  • why existing practice appears sufficient;
  • the failure mechanism;
  • what existing systems solve and leave open;
  • the proposed architecture or decision;
  • evidence and competing explanations;
  • current gaps;
  • promising directions;
  • evaluation and evidence that would change the conclusion.

Do not front-load a source matrix, equation, related-work parade, or newly named property.

6. Write The Required Gap Section

Use a reader-facing heading such as:

  • ## Where current work is still weak
  • ## 现有研究还缺什么

Cover two to five concrete gaps. For each gap, explain:

  • what leading papers or systems already achieve;
  • the exact missing capability or evidence;
  • why the omission matters to correctness, performance, security, operation, cost, or adoption;
  • what observation or experiment would establish whether the gap is material.

Do not repeat individual papers' future-work paragraphs. Synthesize across evidence.

7. Write The Required Ideas Section

Use a heading such as:

  • ## Promising directions with academic and production value
  • ## 兼具学术价值与生产价值的方向

Prefer two or three developed ideas over a brainstorm list. Every idea must include:

  1. Gap: the missing mechanism or evidence it addresses.
  2. Mechanism: what the proposed system, abstraction, dataset, protocol, or algorithm actually does.
  3. Delta: how it differs from the strongest adjacent work.
  4. Artifact: what can be implemented, released, or reproduced.
  5. Evaluation: workloads, baselines, metrics, and an ablation or counterexample that distinguishes the idea.
  6. Academic value: the generalizable question, property, or method.
  7. Production value: who can deploy it and which cost or failure it reduces.
  8. Failure condition: a result showing the extra mechanism is not worth its complexity.

Reject ideas that are only "apply an LLM," "build a platform," "add eBPF," or "use a better model." The mechanism must remain meaningful when the label is removed.

8. Write Bilingual Pages Naturally

English and Chinese versions share evidence, claims, and idea quality, not sentence boundaries.

In Chinese:

  • use Chinese for ordinary concepts;
  • retain English only for proper nouns, identifiers, code, and useful terms of art;
  • do not make the grammar depend on a stack of English nouns;
  • explain a formal term in Chinese before relying on its English name.

9. Use Reader-Facing Conclusion Boundaries

Do not publish headings such as Scope, limitations, and falsification or direct Chinese translations of that template.

Use a natural heading such as:

  • ## What would change this conclusion?
  • ## 哪些结果会改变这个判断?

Explain assumptions, counterexamples, and decisive future evidence in prose. Keep the concepts, remove the template language.

10. Validate And Publish

Every scheduled daily run must finish with one new bilingual report deployed publicly.

  1. create a focused branch;
  2. add exactly one new Daily Report topic in English and Chinese;
  3. change only files in scope plus directly coupled navigation/metadata/operating records;
  4. run the repository's full verification workflow;
  5. inspect the complete PR diff;
  6. merge only after required checks pass;
  7. verify that the exact merge commit deploys;
  8. inspect the public English and Chinese pages, navigation, canonical URLs, gap sections, idea sections, and rendered conclusion heading;
  9. record the report's series and topic classification for the next rolling-mix calculation.

A draft, open PR, green pre-merge check, revision-only day, or unpublished report is not completed daily delivery.

Publication Review

Reject or revise the candidate when any of these are true:

  • the title depends on an unexplained coined term;
  • the opening reads like an abstract rather than a human explanation;
  • the report summarizes papers one by one;
  • the gap section contains generic complaints;
  • the ideas section lacks an implementable artifact or discriminating evaluation;
  • academic value is claimed only because the topic is new;
  • production value is claimed without a deployable boundary or user;
  • repository-owned work is treated more generously than outside work;
  • the page contains process-oriented warning boxes, provenance notes, review-status text, or footer disclaimers;
  • the new page substantially duplicates an existing question or thesis;
  • important claims cannot be traced to primary evidence;
  • the topic would violate the rolling eBPF/Agent editorial mix.

Daily Fallback

There is no no-report outcome for the scheduled daily operation.

If the initial candidate lacks a defensible gap, mechanism, evidence base, or non-trivial idea:

  1. preserve any useful evidence internally;
  2. reject the candidate;
  3. move to the next question in the active eBPF series;
  4. if necessary, move to another approved eBPF or adjacent-systems series while respecting the rolling mix;
  5. continue until one candidate meets the full publication standard.

The fallback is broader research, not weaker writing.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.