agentleFS
NVIDIAKnown publisherNVIDIA/hyperflare-inference/.agents/skills/repo-prose/SKILL.md

repo-prose

Skill807 starsChanged 4 days agoApache-2.0View on GitHub

Write or review repository prose and technical reports that make sense without conversation or development history. Apply to documentation, code prose, commits, PRs, issue comments, plans, and agent reports.

2 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions2 projects use a version of this.

    Word for word · 2

    1. NVIDIA/flare-local-inference★ 807100% same
    2. NVIDIA/spark-local-inference★ 807100% same

    Edited · 0

    None found.

    Wrote their own version · 0

    None found.

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository2 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    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:

    All repository prose must make sense to a technically capable reader who has the repository but none of the conversation or development …

    line 8
  2. Very specific:

    Do not use lifecycle labels as identities: Phase 2, pilot, next, current, new, old, latest, and similar terms are not technical names.

    line 15
  3. Very specific:

    Do not use false definite references.

    line 17
  4. Distinctive:

    Do not add commentary about why a statement is present or how the author wants the reader to interpret it.

    line 45
  5. Distinctive:

    Read the final text as a reader without the development conversation.

    line 50
  6. Distinctive:

    When reporting a prose defect, quote the passage, identify the missing context or unsupported claim, and describe the information needed to …

    line 56
  7. Distinctive:

    Use these forms only when the implementation and evidence support the statements.

    line 69
  8. Distinctive:

    Use ordinary readable sentences.

    line 84

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

Write Without Hidden Context

All repository prose must make sense to a technically capable reader who has the repository but none of the conversation or development history. Apply this skill to documentation, comments, docstrings, names, errors, commit and PR text, issue and review comments, plans, progress messages, handoffs, and completion reports.

  • Describe the system as it exists. State its purpose, behavior, invariants, interfaces, evidence, and limitations directly. Do not narrate the journey.
  • Do not use lifecycle labels as identities: Phase 2, pilot, next, current, new, old, latest, and similar terms are not technical names.
  • Do not use false definite references. Phrases such as the 1M-token capture, the experiment, or this approach require a locally introduced, unambiguous referent. Counts, dates, and versions are attributes, not identities.
  • On first reference, give the object's semantic role and, when relevant, its durable identifier: artifact name, path, schema, revision, manifest, or hash.
  • Explain concepts before identifiers. Do not make internal codenames, experiment labels, profile numbers, implementation shorthand, or names such as XOR-Cheb-T12 the vocabulary of the design. Mention literal identifiers only after describing what they mean and only when the reader needs them.
  • Documentation is a present-state specification, not a changelog. Replace stale claims instead of layering history on top. The repository keeps no historical documents: do not write chronology, rejected attempts, retrospectives, defect reports, or run journals into any document.
  • Distinguish implementation from qualification. Where support or readiness matters, state what is implemented and the conditions under which it is qualified; identify research-only or unsupported behavior explicitly. An implementation or passing test suite does not establish support for untested workloads.
  • State evidence as conditions, measurement, result, and conclusion. Identify the tested artifact and relevant configuration, separate measurements from hypotheses, and disclose missing hardware, inputs, and unrun checks.
  • Comments explain invariants, intent, and non-obvious constraints, never change history. TODOs name the missing condition and the criterion for removal.
  • Commits and PRs state the problem, resulting behavior, technical reason, compatibility impact, and validation relevant to the change. They do not recount attempts or pivots. Scale detail to the change; a small correction does not need a full report.
  • State the substantive fact, requirement, assumption, or limitation directly. Do not add commentary about why a statement is present or how the author wants the reader to interpret it.

Review before publishing

Read the final text as a reader without the development conversation. Check that the affected component, behavior, and evidence can be understood from the text and its explicit references. Rewrite unexplained references and claims that exceed the evidence before committing, posting, or sending the result. Review revised PR titles and descriptions as well as the text accompanying code changes.

When reporting a prose defect, quote the passage, identify the missing context or unsupported claim, and describe the information needed to repair it. Supply a rewrite when the facts are known; do not invent details to make it sound complete. Judge meaning, not the presence of individual words on a banned list.

Illustrative rewrites:

Requires hidden contextStates the behavior or evidence
"The new path fixes the capture issue.""Mixed speculative export preserves separate hidden-state and sampling buffers, allowing Qwen graph construction to complete."
"Phase 2 removes the duplicate buffers.""Dense Qwen graph variants share scratch storage because their executions are serialized."
"All tests passed; ready to ship.""The GPU regression suite passed on SM120. Tensor-parallel execution has not been tested."

Use these forms only when the implementation and evidence support the statements.

Final test: if understanding any sentence requires "you had to be there," rewrite it.

Delegation and reports

  • Include this skill's resolved absolute path in each delegated assignment and require the delegate to read and follow it for progress messages, findings, completion reports, and handoffs. Pass the requirement to further delegates.
  • Supply the task context and relevant artifacts explicitly. A skill reference does not replace the context needed to do the work.
  • Lead reports with the result and its practical consequence. Identify the affected component or artifact, provide supporting evidence, and distinguish applied changes from proposals and executed checks from planned checks.
  • Use ordinary readable sentences. Keep spaces around words and numbers; do not compress reports into concatenated tokens or unexplained abbreviations.
  • Resolve ambiguous references and unsupported conclusions in returned reports before incorporating them into documentation, decisions, or published text.

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.