Skip to content

exadel-inc/agentic-readiness-assessment

v4.5.1Apache-2.0

Evaluates a software repository for readiness to be developed and maintained by AI coding agents, and reports an evidence-based readiness scorecard. A second skill audits how much of your agents' run time goes to slow tool calls.

Changelog

All notable changes to this plugin are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

Added

  • A second skill, agent-tool-call-latency. It measures every tool call agents made on the machine from local session history, in any agent client, and ranks fixes for the waste: foreground waits killed at the tool's cap, locks and fixed ports that serialise parallel agents, approval and reviewer overhead, slow client releases, and start-up repeated for every call. It ships adapters for Claude Code, Codex, T3 Code, Cursor and Antigravity, a shared call-record format that adapters for other clients write to, and a hook-based capture for clients whose logs have no per-call timing. An example report from a real run is in examples/agent-tool-call-latency/.
  • A CI job that runs every skill's script tests, and plugin and marketplace descriptions that name the second skill.
  • A closing line at the end of every report, pointing to Exadel's AI Maturity & Readiness Assessment with a tracked link.
  • Automatic v<version> tags and published GitHub releases after validation succeeds on main.
  • A dependency-free script that updates or verifies the version across every platform manifest.
  • A repository script for the GitHub tag and release publication performed by CI.

4.5.1 - 2026-09-23

Added

  • gemini-extension.json, the manifest the Gemini CLI extension gallery indexes.
  • SECURITY.md, with a disclosure channel and a plain statement of what the skill reads, runs and writes on a user's machine.
  • Install instructions for Gemini CLI, GitHub Copilot CLI, JetBrains Junie, npx skills add, and clients that discover skills on the filesystem.
  • Support and privacy sections in the README.
  • A CI workflow that parses every manifest, checks that the mirrored versions agree with plugin.json, and runs claude plugin validate . --strict.
  • A CLA signature workflow for pull request contributors.
  • author, homepage, repository and license on the Claude Code marketplace entry.
  • Gemini and other-client publishing routes in docs/PUBLISHING.md.
  • A HOL Plugin Scanner workflow for the community Codex marketplace requirements, with offline scanning and no automated submissions.
  • The Codex composer icon field and a square Colleague icon for both Codex image fields.
  • Dependabot updates for pinned GitHub Actions.

Changed

  • CONTRIBUTING.md rewritten around what this repository holds. The previous version was scaffolding, referring to a test suite, a linter and an npm script that do not exist here, and ending in two unfilled headings.
  • The pull request template moved from the repository root to .github/, so the plugin payload carries only what a client needs.
  • One product name that survived the first redaction pass generalized in the published example report, and the redaction note in examples/README.md corrected to match.
  • The Exadel services pitch appears once, at the end of the README, instead of twice.
  • GitHub Actions in the validation workflows are pinned to immutable commits.
  • The marketplace scanner uses the public-marketplace profile and runs the Cisco deep skill scan.
  • Platform manifest versions are synchronized at 4.5.1. The assessment prompt and scoring contract are unchanged.
  • The README header image is named banner.png; logo.png contains the square Colleague icon.

Removed

  • The README line saying the marketplace catalog had not yet reached the default branch. It reached it in 4.1.1.

4.5.0 - 2026-09-07

Added

  • Dependency-free deterministic validation of the generated report contract.
  • The assessment delegation and decision record, covering deterministic calculations, report validation and human approval boundaries.

Changed

  • Host execution evidence is now required for model identity, permissions, command boundaries, stop conditions and context telemetry.
  • Assessment instructions are model-neutral, portable from an installed skill directory, and explicit about two-phase report finalization.
  • Report validation strengthened for status conditions, negative controls, Gate 3 anchor agreement and escaped Markdown table pipes.
  • The version moved to 4.5.0.

4.1.1 - 2026-09-02

Added

  • Initial plugin scaffold with the agentic-readiness-assessment skill.
  • Assessment prompt imported from its source repository.
  • Example report under examples/, from a run against a Next.js application, linked from the README.
  • Native Codex and Claude Code plugin manifests.
  • A self-hosted Claude Code marketplace catalog and publication guide.

Removed

  • screenshot.png, an unused placeholder carried over from the repository template.

Changed

  • The version in plugin.json is the canonical source of truth for the skill and is mirrored into the platform manifests. The skill reads the root manifest and records it as prompt_version in generated reports. See Versioning.