agentleFS
harlan-zwharlan-zw/brundlefly/skills/technical-guide/SKILL.md

technical-guide

Skill7 starsChanged 32 days agoMITView on GitHub

Writes, verifies, or refreshes technical guides: tutorials, how-to guides, and developer documentation, checked against current code and official sources. Use whenever someone asks for a tutorial or how-to, asks to document how developers do a task, or asks to update stale docs, even when the guide looks short.

The file

Show its rules and text

We split this skill into the rules it gives an agent

56 rules. About 0 are common, 56 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈56

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Create a guide or refresh an existing one.

    line 10
  2. Distinctive:

    Read the target's instructions, contribution rules, terminology, style guide, and nearby documentation.

    line 14
  3. Distinctive:

    Use established rules and filenames.

    line 15
  4. Distinctive:

    Read an existing page in full before editing it.

    line 16
  5. Distinctive:

    Ask only when a missing choice changes the procedure or audience materially.

    line 21
  6. Distinctive:

    Keep one primary purpose. Link deeper background where it would interrupt the steps.

    line 35
  7. Distinctive:

    Read guide content and formats and ensure the reader receives each applicable element.

    line 36
  8. Distinctive:

    Treat these as a completeness check, rather than mandatory headings or a fixed page length.

    line 37
  9. Distinctive:

    Follow the guide format selection procedure in that reference before proposing additional work.

    line 40
  10. Distinctive:

    Read content boundaries when the page mixes guidance, comparison, research, or personal experience.

    line 41
  11. Distinctive:

    Read evidence and verification before adding technical claims or executing examples.

    line 49
  12. Distinctive:

    Use official documentation for public support and platform guarantees.

    line 50
  13. Distinctive:

    Use the matching source revision, exports, tests, and CLI help for implementation details.

    line 51
  14. Distinctive:

    Keep working evidence outside published collections and generated public outputs.

    line 56
  15. Distinctive:

    Do not require four new policy files for a single page.

    line 57
  16. Distinctive:

    If evidence conflicts, narrow the claim or report the missing decision.

    line 61
  17. Distinctive:

    Put version limits and consequential conditions beside the affected step.

    line 70
  18. Distinctive:

    Put supported warnings and failure explanations beside the affected step or claim.

    line 73
  19. Distinctive:

    Do not append generic "Common mistakes", "Best practices", or "Check your understanding" sections.

    line 74
  20. Distinctive:

    Use a separate section only when the brief or reader task needs it.

    line 75
  21. Distinctive:

    Prefer connected prose for explanations.

    line 77
  22. Distinctive:

    Keep items short and lists to five items or fewer by default.

    line 78
  23. Distinctive:

    Use a compact table for parallel rules or a worked example for behavior when it helps comprehension.

    line 81
  24. Distinctive:

    Do not add headings, visuals, or mechanical paragraph breaks merely to vary the page.

    line 82
  25. Distinctive:

    Use the project's language and package manager.

    line 86
  26. Distinctive:

    Use placeholders that readers can recognize and replace.

    line 87
  27. Distinctive:

    Write direct, natural prose.

    line 89
  28. Distinctive:

    Do not use em dashes or hyphens as sentence punctuation.

    line 90
  29. Distinctive:

    Use commas, semicolons, colons, or separate sentences.

    line 91
  30. Distinctive:

    Prefer a concrete explanation to hype or stock transitions.

    line 92
  31. Distinctive:

    Use consistent terms. Leave strong sentences unchanged.

    line 93
  32. Distinctive:

    Use passive voice when the actor is irrelevant or the object deserves emphasis.

    line 94
  33. Distinctive:

    Add diagrams or screenshots when they answer a reader question, rather than to decorate the page.

    line 95
  34. Distinctive:

    Use supplied experience and measurements.

    line 96
  35. Distinctive:

    Never use the mascot or an agent persona to imply human authorship.

    line 98
  36. Distinctive:

    Use exact names and stated pronouns.

    line 99
  37. Distinctive:

    Use clarity for the editorial review.

    line 106
  38. Distinctive:

    After prose edits, apply claim fidelity against the draft and inspected evidence.

    line 107
  39. Distinctive:

    Use the actual documented commands and record their results.

    line 110
  40. Distinctive:

    Read the verification reference for limitations, failures, and publication checks.

    line 112
  41. Distinctive:

    Check relevant conditions and response metadata, rather than only a successful command or visible result.

    line 114
  42. Distinctive:

    If a check fails, repair the example, explanation, or visual, then repeat the affected checks.

    line 116
  43. Distinctive:

    Keep the initial draft and observed failures in private evidence so the repair remains inspectable.

    line 117
  44. Distinctive:

    Technical: check behavior, versions, code, prerequisites, outcomes, and link destinations against evidence.

    line 121
  45. Distinctive:

    Editorial: check task order, understandable concepts, useful detail, and preservation of meaning and voice.

    line 122
  46. Distinctive:

    Keep unavailable checks in verification limits.

    line 127
  47. Distinctive:

    Leave clear passages unchanged.

    line 128
  48. Distinctive:

    Use the guide content and formats reference to distinguish needless headings from useful steps and lookup sections.

    line 132
  49. Distinctive:

    After prose edits, recheck technical meaning and examples.

    line 134
  50. Distinctive:

    Use the target's documented docs checks and preview when applicable.

    line 136
  51. Distinctive:

    Keep evidence notes and review commentary out of the public article.

    line 144
  52. Distinctive:

    If a claim blocks the core procedure, deliver a draft with the blocker identified in the handoff.

    line 145
  53. Distinctive:

    Do not label that procedure verified.

    line 146
  54. Distinctive:

    If publication is authorized, check contribution rules for restrictions on AI-authored content.

    line 150
  55. Distinctive:

    If those rules prohibit the contribution, stop publication and tell the operator.

    line 151
  56. Distinctive:

    Ask before publication if either is unknown.

    line 153

≈ “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

Technical guide

Help a reader complete or understand a technical task with accurate information and useful examples. Create a guide or refresh an existing one. Keep each change within the requested scope.

1. Establish the reader's task

Read the target's instructions, contribution rules, terminology, style guide, and nearby documentation. Use established rules and filenames. A missing editorial framework does not require creating one. Read an existing page in full before editing it.

Identify the reader, starting knowledge, desired outcome, supported version, and destination. Respect their time and knowledge. Do not explain prerequisites they already meet or blame them for unclear instructions. Infer these from the request and repository when clear. Ask only when a missing choice changes the procedure or audience materially.

Choose the structure that serves the task:

Predict the attention this reader will give the task. Use it to choose scope and depth before drafting. Spend detail on the steps or concepts they need most. Keep complete prerequisites and warnings beside their use.

Reader needsStructure
Learn by doingTutorial with a visible result after each stage
Complete a known taskHow-to guide with prerequisites and ordered steps
Look up exact behaviorReference organized around public names
Understand a designExplanation with supported mechanisms and trade-offs

Keep one primary purpose. Link deeper background where it would interrupt the steps. Read guide content and formats and ensure the reader receives each applicable element. Treat these as a completeness check, rather than mandatory headings or a fixed page length. During the initial sweep, identify which code examples, diagrams, captures, and supporting sources the destination can use. Record what exists, what needs creation, and what cannot be verified in this context. Follow the guide format selection procedure in that reference before proposing additional work. Read content boundaries when the page mixes guidance, comparison, research, or personal experience. For a refresh, prefer the smallest change that resolves the reader's problem. Preserve useful examples, routes, anchors, and deliberate voice.

Done: the reader's goal, version, page scope, and constraints are clear.

2. Establish evidence before drafting

Read evidence and verification before adding technical claims or executing examples. Use official documentation for public support and platform guarantees. Use the matching source revision, exports, tests, and CLI help for implementation details. Reconcile differences between the checkout and the release the reader can install.

Record the source and version behind material claims in existing project records or small private evidence notes. The evidence notes need the claim, evidence location, scope, and any unresolved question. Keep working evidence outside published collections and generated public outputs. Do not require four new policy files for a single page.

Issues and support reports explain reader problems. Verify the suggested remedy against supported behavior. Plans describe intent. Announcements describe release history. Neither proves current behavior alone. If evidence conflicts, narrow the claim or report the missing decision.

Done: the supported path and its material claims have inspected evidence.

3. Draft for the next action

Lead with the outcome and essential prerequisites. Show the common path before alternatives. Introduce a concept before the reader must use it, unless it is an explicit prerequisite. Put version limits and consequential conditions beside the affected step.

For each step, explain what to do and what success looks like. Put supported warnings and failure explanations beside the affected step or claim. Do not append generic "Common mistakes", "Best practices", or "Check your understanding" sections. Use a separate section only when the brief or reader task needs it. Let the narrative carry the explanation and result. Do not add a quiz or advice appendix for completeness. Prefer connected prose for explanations. Use lists when scanning parallel items or following ordered actions helps the reader. Keep items short and lists to five items or fewer by default. Watch for successive dense paragraphs after removing lists. Give each paragraph one clear job and cut repeated explanations. Lead with the rule needed now. Keep consequential conditions beside their use; link deeper background where appropriate. Use a compact table for parallel rules or a worked example for behavior when it helps comprehension. Do not add headings, visuals, or mechanical paragraph breaks merely to vary the page. Group longer procedures into meaningful phases without omitting steps. Preserve required templates and complete reference entries. Provide complete imports, setup, configuration, and working-directory context for runnable examples. Distinguish runnable code from pseudocode and schematic output. Use the project's language and package manager. Preserve exact identifiers and units. Use placeholders that readers can recognize and replace. Keep credentials out of examples.

Write direct, natural prose. Keep useful qualifications, uncertainty, and attribution. Do not use em dashes or hyphens as sentence punctuation. Use commas, semicolons, colons, or separate sentences. Keep this rule during editorial edits. Prefer a concrete explanation to hype or stock transitions. Use consistent terms. Leave strong sentences unchanged. Use passive voice when the actor is irrelevant or the object deserves emphasis. Add diagrams or screenshots when they answer a reader question, rather than to decorate the page. Use supplied experience and measurements. Missing evidence stays missing; never invent it for personality. Preserve the author's supplied voice without claiming their experience as the agent's own. Never use the mascot or an agent persona to imply human authorship. Use exact names and stated pronouns; otherwise use a username or singular they.

Done: the draft covers the supported task without assuming hidden setup.

4. Verify the reader's path

For runnable examples, apply verify examples. Use clarity for the editorial review. After prose edits, apply claim fidelity against the draft and inspected evidence.

For procedural content, run safe examples from the documented starting state when the environment permits. Use the actual documented commands and record their results. For state-changing steps, respect the user's authorization and use an isolated environment where possible. Read the verification reference for limitations, failures, and publication checks. Turn material claims and promised outcomes into concrete checks before calling the guide verified. Check relevant conditions and response metadata, rather than only a successful command or visible result. Render diagrams with the destination's renderer when available. Inspect actual captures and their captions. If a check fails, repair the example, explanation, or visual, then repeat the affected checks. Keep the initial draft and observed failures in private evidence so the repair remains inspectable.

Perform two distinct reviews:

  1. Technical: check behavior, versions, code, prerequisites, outcomes, and link destinations against evidence.
  2. Editorial: check task order, understandable concepts, useful detail, and preservation of meaning and voice.

Report a finding only when evidence shows a contradiction, missing prerequisite, or concrete reading problem within the requested scope. Accept prerequisites and working-directory context supplied by the guide, request, or documented starting state. A supported runtime need not be the earliest compatible runtime unless the guide claims that boundary. Keep unavailable checks in verification limits. Missing verification alone does not prove a defect. Leave clear passages unchanged. Do not request background, recovery advice, or extra success signals without a demonstrated reader need.

Review heading-to-paragraph density for fragmented explanations. Merge unnecessary divisions without padding prose or removing useful navigation. Also inspect successive long paragraphs for buried actions, stacked exceptions, and repetition. Preserve technical conditions during any compression. Use the guide content and formats reference to distinguish needless headings from useful steps and lookup sections.

After prose edits, recheck technical meaning and examples. For a refresh, search adjacent pages for repeated claims and links to changed anchors. Use the target's documented docs checks and preview when applicable.

Done: material claims are supported; repairs pass their affected checks; remaining limits appear in the handoff.

5. Deliver

Provide the guide or requested file changes, then a short evidence handoff. State the version checked, meaningful verification, and unresolved paths. Keep evidence notes and review commentary out of the public article. If a claim blocks the core procedure, deliver a draft with the blocker identified in the handoff. Do not label that procedure verified.

Repository edits, publication, deployment, and external messages follow the user's requested delivery scope. Writing a guide does not grant permission to publish it or operate a production service. If publication is authorized, check contribution rules for restrictions on AI-authored content. If those rules prohibit the contribution, stop publication and tell the operator. Include an early plain authorship sentence naming the operator and actual human review state in a public artifact. Ask before publication if either is unknown. Never claim review from an intention to review later.

Background

The structure uses the reader needs described by Diátaxis. For general developer prose, consult Google's voice and tone guidance. Local editorial rules take precedence over these defaults.

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.