agentleFS
shinprshinpr/ai-coding-project-boilerplate/.claude/skills-en/llm-friendly-context/SKILL.md

llm-friendly-context

Skill239 starsChanged 3 days agoMITView on GitHub

Clarifies inputs, outputs, success criteria, decisions, and unresolved conditions so downstream agents can execute without guessing. Use when writing or revising LLM-facing prompts, handoffs, planning artifacts, reviews, reports, or generated instructions.

2 other projects use a version of this, 2 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 8 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Keep a prohibition only when it protects an irreversible boundary or a shipped contract.

    line 18
  2. Distinctive:

    Prefer concrete file paths and section hints over broad module names

    line 31
  3. Distinctive:

    Follow references while they can change an in-scope decision, action, or verification result.

    line 32
  4. Distinctive:

    Add only constraints that reduce ambiguity or preserve a real requirement

    line 44
  5. Distinctive:

    Keep simple downstream tasks lightweight when the target action, context, and success criteria are already clear

    line 45
  6. Very specific:

    Treat a stated size expectation — minimal, a few lines, an explicit line or file estimate — as one budget over the whole completed diff, …

    line 46
  7. Distinctive:

    Use these rewrites before treating a prompt, handoff, or artifact as complete.

    line 50
  8. Distinctive:

    Before sending a prompt or artifact to another agent, verify:

    line 66
  9. Distinctive:

    Before writing or finalizing a generated document:

    line 80

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

LLM-Friendly Context

The goal is stable downstream execution: the next agent should know what to read, what to do, what counts as success, and which unresolved decisions can change the result.

This skill governs the clarity of LLM-facing output — prompts, handoffs, and generated artifacts. The caller supplies the artifact type and any artifact-specific template or input contract; include only the information its consumer uses to decide, act, or verify. Use a declared contract's field names and value meanings when the consumer branches on them.

Core Rules

  1. Use positive, executable instructions
    • State what the next agent should do
    • Convert quality policies into positive criteria
    • Example: "Preserve existing public API behavior across the documented compatibility cases."
    • Keep a prohibition only when it protects an irreversible boundary or a shipped contract; then name the protected condition and the allowed action alongside it
  2. Make vague instructions concrete
    • Clarify wording when plausible interpretations would change the outcome, scope, authority, or proof. Replace such terms with observable conditions, paths, commands, schemas, or decision rules; for reversible choices, supply the purpose, evidence, and selection criteria instead of a fixed selection
    • Terms that often need clarification when they leave a decision to the next agent: appropriate, proper, related, existing behavior, optional, as needed, if needed, per convention, unresolved alternatives, TBD, placeholder
  3. Specify output shape
    • Define the sections, fields, table columns, JSON keys, or checklist items the consumer uses
    • For handoffs, include produced artifact paths and status fields only when they control the next transition
  4. Provide the smallest sufficient context
    • Sufficient means sufficient for the assigned action, not complete background: every included item is consumed by that action or by the result it must produce
    • Include the purpose, source artifacts, hard constraints, accepted decisions, and unresolved conditions that action consumes
    • Prefer concrete file paths and section hints over broad module names
    • Follow references while they can change an in-scope decision, action, or verification result; stop when the next link only confirms what is already decided
  5. Decompose complex work into verifiable steps
    • Expose real dependencies as ordered steps; let independent reversible work follow the route suited to the task
    • Each step needs a checkpoint: what evidence proves it is complete
  6. Permit uncertainty explicitly
    • Resolve missing operational detail from governing artifacts and representative repository evidence before treating it as unresolved
    • Record remaining uncertainty with its effect on the outcome or proof. Make reversible repository-local choices inside the value boundary and preserve the evidence used
    • Route an unknown that blocks the next step as an exact evidence prerequisite. Ask the user only when a value-boundary choice belongs to the user, or when an irreversible external action requires authorization. When only proof is unavailable, complete unaffected work and report exactly what could not be verified and why
  7. Keep constraints proportionate
    • Add only constraints that reduce ambiguity or preserve a real requirement
    • Keep simple downstream tasks lightweight when the target action, context, and success criteria are already clear
    • Treat a stated size expectation — minimal, a few lines, an explicit line or file estimate — as one budget over the whole completed diff, not per file or per step. When the work cannot fit it, report the overrun and the reason instead of silently exceeding it

Rewrite Patterns

Use these rewrites before treating a prompt, handoff, or artifact as complete.

Ambiguous formRewrite as
optional used as an unresolved choiceRequired, omitted, or required only under a named condition
Alternatives whose choice changes a protected outcome, contract, or authorityThe governing constraint, the selected option, or the unresolved owner decision; for reversible choices, the purpose, evidence, and selection criteria
as needed / if neededThe triggering condition and required action
per conventionThe file, function, test, or documented convention to follow
related filesSpecific paths, globs, or search hints
existing behaviorThe observable behavior, source file, test, API response, or UI state to preserve
placeholderExact temporary value/behavior, allowed dependencies, and verification expectation
TBD used as a placeholder for required informationThe decision it can change and the exact evidence prerequisite; omit it when the item has no downstream effect
appropriate / properA measurable criterion or checklist

Handoff Checklist

Before sending a prompt or artifact to another agent, verify:

  • The target action is explicit
  • Required input paths, source artifacts, and decision-relevant facts are named
  • Every included context item is consumed by the target action or its required result
  • Accepted decisions and constraints are stated once, without alternate wording
  • Output format or expected status fields are specified
  • Success criteria are observable
  • Outcome-relevant ambiguities are resolved or assigned to their decision owner; harmless execution flexibility remains
  • Any stated size expectation is expressed as one budget over the completed diff, with the overrun-reporting condition named
  • The next agent can complete its scope from the supplied purpose, sources, criteria, and evidence, or return one exact evidence prerequisite or authoritative workflow stop

Generated Artifact Checklist

Before writing or finalizing a generated document:

  • Each requirement, claim, task, test skeleton, or review finding has enough source context to trace why it exists
  • Every executable instruction names the target, action, and expected result
  • Verification steps say what to run or observe and what result proves success
  • If an artifact is derived from another artifact, copied decisions stay consistent in wording and meaning
  • Any stated size expectation is expressed as one budget over the completed artifact, with the overrun-reporting condition named
  • Missing information records the decision or proof it affects; only a user-owned value-boundary choice or irreversible external action authorization is a blocking escalation

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.