Cachely
Cachely Doctor

Is your repository actually cacheable?

Give your coding agent one prompt. It loads Cachely's audit skill, reviews your Nx or Turborepo configuration in its existing repository context, and returns evidence-backed findings without sending your repository to Cachely.

Run the local audit

Paste this prompt into a coding agent that can read the repository you want to inspect:

Goal: Audit this repository for build-cache readiness and reduce repeated build, test, lint, and generation work.

Issue: Tasks may be missing the cache because of incorrect inputs or outputs, undeclared environment variables, nondeterministic values, inconsistent toolchain versions, or unsafe cache configuration.

Fix: Run a read-only Cachely cache audit inside this coding agent. Identify the build system and cache provider, inspect cacheable tasks, and report evidence-backed correctness, security, and performance issues. Do not modify files, execute project scripts, send repository content or findings to Cachely, or estimate cache-hit rates or savings without runtime data.

Skill: https://cachely.dev/.well-known/agent-skills/cachely-cache-audit/SKILL.md

Docs: https://cachely.dev/cache-audit, https://cachely.dev/improve-build-cache-hit-rate

The canonical skill is published at https://cachely.dev/.well-known/agent-skills/cachely-cache-audit/SKILL.md. Agents fetch the current instructions directly from Cachely, so the short prompt stays stable as the audit improves.

What Cachely checks

  • Nx and Turborepo task inputs, outputs, dependencies, and cache settings.
  • Undeclared environment variables and local-versus-CI toolchain drift.
  • Timestamps, Git metadata, temporary paths, randomness, and other nondeterministic output sources.
  • Over-broad inputs, missing outputs, coarse tasks, and fragmented cache namespaces.
  • Write-capable credentials exposed to untrusted pull requests or forks.

The first version provides deep Nx and Turborepo coverage. It detects Gradle and Bazel repositories but labels their assessment as limited rather than overstating confidence.

What the report contains

Every finding includes severity, confidence, the affected task, file-level evidence, impact, a minimal remediation, the risk of that change, and a verification step. Tasks are classified as safe, probably safe, unsafe, or lacking enough evidence.

The Cache Readiness Score measures static configuration quality. It is not a measured cache-hit rate and does not invent execution-time or financial savings. For measured behavior, connect the repository to Cachely and compare real cache operations.

Privacy and usage measurement

The audit is read-only and makes no audit-initiated network requests after the skill is loaded. It does not execute project scripts, read secret values, or send files or the generated report to Cachely.

Cachely may count aggregate requests for the public SKILL.md file through ordinary edge request analytics. That tells us the skill was fetched, not that an audit completed. No repository identifier, path, source content, report data, or persistent machine identifier is collected.

Your coding-agent provider may process repository data under its own privacy terms. Cachely Doctor does not change that provider relationship.

To investigate an observed miss after the static audit, continue with the cache-hit-rate troubleshooting guide.