agentleFS
NVIDIAKnown publisherNVIDIA/TensorRT-Edge-LLM/skills/edge-llm-documentation/SKILL.md

edge-llm-documentation

Skill631 starsChanged 3 days agoApache-2.0View on GitHub

Write, edit, or review TensorRT Edge-LLM documentation using the Google developer documentation style guide. Apply to READMEs, tutorials, how-to guides, conceptual and design documents, API and CLI references, and release notes; not ordinary code changes or chat replies.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 9 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    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:

    Use the Google developer documentation style guide as the editorial reference for the whole document, including organization, language, …

    line 14
  2. Distinctive:

    Read the bundled Google documentation practices before writing or reviewing.

    line 17
  3. Distinctive:

    Read the requested document, nearby pages, and applicable repository instructions.

    line 24
  4. Distinctive:

    Use the user's explicit requirements first, then explicit project conventions, then Google's guide.

    line 26
  5. Distinctive:

    If a needed source cannot be accessed, state the verification gap rather than inventing a Google rule.

    line 30
  6. Distinctive:

    Use Markdown for new prose documents unless the user or destination's conventions specify another format.

    line 52
  7. Very specific:

    Read docs/README.md and docs/source/conf.py before changing site syntax or building the site.

    line 62
  8. Very specific:

    Follow CODING_GUIDELINES.md for code examples and Python docstring syntax.

    line 69
  9. Distinctive:

    Verify commands, flags, paths, defaults, and dependencies against the target branch.

    line 71
  10. Distinctive:

    Keep documented benchmark results attached to their model, hardware, precision, workload, and measurement conditions.

    line 75
  11. Distinctive:

    If an external edit needs confirmation, give a brief confirmation or artifact link without a recap.

    line 109

≈ “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
<!-- SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. --><!-- SPDX-License-Identifier: Apache-2.0 -->

Edge-LLM documentation

Use the Google developer documentation style guide as the editorial reference for the whole document, including organization, language, formatting, links, accessibility, and technical examples. Read the bundled Google documentation practices before writing or reviewing; this summary is sufficient for routine work. Consult individual Google pages only for uncovered topics or ambiguities; there is no need to reread the entire guide for each task.

Establish the scope and sources

Read the requested document, nearby pages, and applicable repository instructions. Identify the audience, reader's task, target release or branch, and output format. Use the user's explicit requirements first, then explicit project conventions, then Google's guide. Existing prose can contain mistakes; do not treat every existing spelling or layout as a required convention.

If a needed source cannot be accessed, state the verification gap rather than inventing a Google rule. Do not apply Android- or Google Cloud-specific conventions to Edge-LLM without a relevant reason.

For product facts, inspect branch-local implementation, command help, build configuration, tests, and maintained documentation. Google's guide determines editorial choices, not supported models, platforms, defaults, or behavior.

Write or revise

Choose a structure suited to the request: a learning sequence for a tutorial, an actionable procedure for a how-to, an explanation for a concept, or precise lookup information for a reference. Keep a narrow correction narrow. For a review-only request, report findings without editing files.

Apply the reference across the requested scope. Preserve technical meaning, literal identifiers, and prerequisites while improving prose. Explain proposed behavior as proposed in design documents; describe implemented behavior in user guides. Do not convert every document into a tutorial or force a fixed template.

Apply Edge-LLM conventions

  • Use Markdown for new prose documents unless the user or destination's conventions specify another format. Use Jupyter notebooks only for explicitly requested executable tutorials.
  • Preserve the document's Markdown, reStructuredText, or AsciiDoc format and supported directives. Keep Sphinx cross-references, explicit anchors, and navigation working when headings or paths change.
  • For work under design/, read the design index and design contribution guidance. Follow their document structure and AsciiDoc conventions, and verify technical claims against current branch code and tests.
  • Read docs/README.md and docs/source/conf.py before changing site syntax or building the site. Use the existing renderer's code blocks and admonitions; do not transplant Google's website-specific HTML or CSS.
  • Identify generated documentation before editing. For generated API pages, update the owning Python docstring, Doxygen comment, or generator, then regenerate the output. Direct edits to generated pages can disappear during the next documentation build.
  • Follow CODING_GUIDELINES.md for code examples and Python docstring syntax. Google's documentation prose guidance does not replace C++ or Python rules.
  • Verify commands, flags, paths, defaults, and dependencies against the target branch. Separate host setup from device execution and identify relevant platform, precision, and version constraints. Do not combine distinct export or runtime routes into an apparently interchangeable procedure.
  • Keep documented benchmark results attached to their model, hardware, precision, workload, and measurement conditions. Do not turn an observed result into a general performance or support guarantee.
  • Public documentation must be usable from the public tree. Preserve release exclusion markers and avoid introducing internal-only links or prerequisites into public instructions. Link to a maintained procedure instead of copying commands that can diverge.

Verify and deliver

Review factual correctness separately from editorial quality. Check local links, anchors, navigation, code block syntax, and placeholders. For site changes, run the documented build when dependencies are available and inspect affected rendered pages. Run applicable pre-commit hooks on changed files.

Validate changed examples when the environment permits. Distinguish static checks from execution; a documentation build does not prove that a model works. A model-validation claim requires export, engine build, and inference under the repository's validation policy.

For a style review, give the location, reader impact, suggested correction, and the relevant Google source or project convention. Distinguish factual defects from editorial suggestions; preserve justified exceptions that improve clarity.

Apply edits directly to the target document. Write for its readers without recapping the prompt, conversation, or editing process. Do not add sections that summarize or justify the editing process unless explicitly requested. Keep content the document's reader needs, such as architectural rationale or steps to verify a tutorial's result. Preserve useful technical context and attribution; omit claims about excluded approaches that merely repeat a conversational constraint. Do not remove existing sections or restructure unrelated content solely to apply this guidance.

For editing requests, return only requested file changes or a commit message. If an external edit needs confirmation, give a brief confirmation or artifact link without a recap. Still perform relevant checks and disclose actual blockers or failed actions.

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.