Combined AI-Readiness

Run the bundled evaluator for the host operating system once. The executable owns repository inspection, evidence collection, Copilot judge consensus, scoring, conservative execution-aware documentation drift analysis, report generation, and CSV export; do not reproduce those stages manually.

For questions about how the evaluation is designed or how scores are calculated, read and cite how-it-work.md. Do not run an evaluation merely to explain the methodology.

Workflow

1. Determine the target directory:
   • Use the directory supplied by the user when one is provided.
   • Otherwise use the user's current working directory.
   • Resolve it to the repository root with git -C <target> rev-parse --show-toplevel.
   • Stop with the Git error if the target is not inside a Git repository.

2. Detect the host operating system and architecture, then resolve the matching executable beside this SKILL.md:

   • On Windows amd64, use ai-readiness-eval.exe.
   • On Linux amd64 (including Ubuntu x86_64), use ai-readiness-eval-linux-amd64.
   • In an installed plugin, resolve the selected filename under ${CLAUDE_PLUGIN_ROOT}/skills/codeblend-ai-composite/.
   • In the source lab, resolve it in this skill directory.
   • On Linux, run chmod +x on the selected executable before invoking it in case the plugin installation did not preserve executable permissions.
   • Stop clearly on macOS, Linux ARM64, or another unsupported platform; do not try to run the Windows executable through Wine.
   • Stop clearly if the executable is missing; do not fall back to the legacy Python composite pipeline or scanner.

   On Windows, require [Runtime.InteropServices.RuntimeInformation]::OSArchitecture to be X64. On Linux, require uname -m to return x86_64 or amd64. Use the actual host result rather than inferring the platform from repository paths or the user's shell syntax.

3. Run the evaluation against the resolved repository root:

   Windows PowerShell:

   & $Evaluator eval $RepoRoot

   Linux shell:

   "$evaluator" eval "$repo_root"

   Keep the user's current working directory unchanged. Pass through any evaluation flags the user explicitly supplied, such as --models, --max-rounds, --allow-missing-api-evidence, --full-findings, or --no-cache.

4. Read the run directory printed by the CLI, then present its composite.md. Lead with the headline score, AI-ready verdict, substrate level, operation tier, highest-priority remediation items, and documentation drift coverage and residual action. Treat semantic review as advisory and automatic repair as a backstop; only a deterministic repository-wide blocking PR gate is complete coverage. Surface the continuous-cleanup automation rating when reported; the evaluator refines generic scheduled cleanup guidance only when executable cleanup capability is independently proven. Also report the run directory so the user can access the JSON and CSV artifacts.

Runtime requirements

• git and GitHub Copilot CLI must be available on PATH.
• GitHub repositories require an authenticated gh session for API evidence.
• Azure DevOps repositories require an authenticated az session.
• Do not add --allow-missing-api-evidence unless the user explicitly accepts a degraded evaluation.

Feedback

Report skill issues at https://github.com/gim-home/codeblend/issues.