agentleFS
divyanshu-iitiandivyanshu-iitian/agent-website-design-skills/agentic-site-building/SKILL.md

agentic-site-building

Skill2 starsChanged 4 months agoMITView on GitHub

Design, implement, or harden an AI website-building agent that turns natural-language requests into scoped code changes with planning, structured edits, deterministic verification, bounded repair, safe previewing, version history, and token-aware model routing.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    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:

    Prefer a structured patch for an existing artifact.

    line 14
  2. Distinctive:

    If validation fails, give the verifier evidence to one bounded repair attempt.

    line 16
  3. Distinctive:

    Keep generated code facts grounded in user data.

    line 30
  4. Distinctive:

    Read references/orchestration-contract.md when defining schemas, retry policy, or production metrics.

    line 45

≈ “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

Agentic Site Building

Build editing systems that can prove a requested change happened without rewriting unrelated work or claiming success on a no-op.

Workflow

  1. Map the editable artifact, source of truth, preview boundary, model providers, persistence, and rollback path.
  2. Compile each request into an explicit contract: mode, target surface, allowed change radius, acceptance criteria, and risk.
  3. Prefer a structured patch for an existing artifact. Require exact, unique anchors and a bounded change budget.
  4. Apply patches locally, then validate syntax, structure, requested-surface change, and collateral-change limits.
  5. If validation fails, give the verifier evidence to one bounded repair attempt. Escalate to a full rewrite only when scope requires it or patching is unsafe.
  6. For redesigns, reject the current composition deliberately while preserving factual content and product constraints.
  7. Render generated code in an isolated preview. Keep authentication, storage, parent DOM, and privileged network access outside the generated runtime.
  8. Save only accepted outputs. Keep restorable revisions with prompt, mode, verifier result, and prompt/model version.
  9. Expose honest progress and failure states. Never label an unchanged or unverified preview as updated.
  10. Measure patch success, no-op rate, repair rate, latency, token cost, rollback rate, and user acceptance.

Decision Rules

  • Deterministic code owns safety and common runtime contracts; the model owns interpretation and creative composition.
  • A small edit should not receive the full site unless the patch path fails safely.
  • Verification must inspect the requested surface, not only whole-file similarity.
  • Retries must use new evidence and stop after a fixed budget.
  • Provider fallback handles availability; it must not silently weaken required structured-output or context capabilities.
  • Keep generated code facts grounded in user data. Do not invent projects, metrics, employers, or media.

Delivery Contract

Provide or implement:

  1. Execution contract - modes, scope, acceptance criteria, and escalation rules.
  2. Orchestration path - plan, patch or build, validate, repair, persist, preview, rollback.
  3. Verifier design - deterministic checks and evidence returned to repair attempts.
  4. Safety boundary - generated-code sandbox and authorization limits.
  5. Efficiency budget - context selection, model routing, retries, and stopping conditions.
  6. Operational proof - tests for no-op, collateral edit, failed patch, retry, rollback, and preview isolation.

Reference

Read references/orchestration-contract.md when defining schemas, retry policy, or production metrics.

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.