Agentic 1.4.0

Agentic 1.4.0 makes installation explicit across twelve coding harnesses and documents the complete engineering workflow, from standards and requirements through design review, implementation, QA, and security.

Installation

• Native Claude plugin metadata, a portable plugin manifest for Codex and Antigravity, a Codex marketplace, and Pi package metadata.
• Named global/project skill targets for Claude, Codex, Pi, OpenCode, Antigravity CLI/App, Gemini CLI, Cursor, GitHub Copilot, Cline, Windsurf, and Amp.
• Dependency-aware selected installs, destination deduplication, local links, and update/removal coverage.
• Project setup now generates AGENTS.md by default and selects the current harness’s instructions.

Documentation and verification

• README covering the full scope of Agentic, with direct installation and first-use instructions.
• Linked installation, usage, language standards, and development/verification guides.
• Registry-generated, version-synchronized manifests and strict validation of the packaged distribution.
• A distributable that retains standalone resources and licenses while excluding development context and local build/cache files.

Updating

For Claude Code:

/plugin marketplace update agentic
/plugin update agentic@agentic

For direct skill installs, repeat the original scope and target with --update, for example:

npx -y github:igmrrf/agentic skills --global --agent codex --update

Explicit --agent all installs to every supported harness location. With no --agent, the default remains Claude plus shared .agents skills. Named global Gemini and Antigravity targets now use their native directories; remove obsolete shared copies only when other harnesses no longer need them.

Full changelog · Installation guide

Agentic 1.3.0

Agentic 1.3.0 adds two independently installable skills within the existing engineering workflow:

• Caveman core: explicitly activated conversational presentation, persistent session mode, truthful status, exact technical payloads, and full artifact/evidence preservation. Proxy, CLI, telemetry, hooks, compression infrastructure, and alternate voices are excluded.
• Security Audit: focused guidance or an explicitly requested six-phase audit, with all phase and attack-class instructions, schema, validators, upstream tests, and Cloudflare MIT notice.

The canonical registry now includes 17 skills. Standalone resources and licenses survive copy/link, archive, npm-package CLI, marketplace, and plain directory installation. Codinary retains delivery ownership, independent QA, review-only scope, its two-fix limit, and action-specific release authorization. Findings do not authorize remediation or replace delivery verdicts.

Pinned sources: Caveman 9b68f7ed12b05e7efaff18e65c5a3fc2849fd0a4 (Apache-2.0 with retained historical MIT notices) and Security Audit c1c8a8c1471069fb0e188eeaff69b8e8db6564a8 (MIT).

Validation includes registry and provenance checks, 127 integration tests, 65 upstream validator tests, 23 installer checks, 10 Codinary recovery tests, package-content inspection, native marketplace validation, and isolated behavioral decision probes. Full native audit orchestration and OS sandbox enforcement remain unverified and fail closed when required controls are unavailable.

Install directly from the GitHub tag:

npx -y github:igmrrf/agentic#v1.3.0 skills --global --skill caveman,security-audit

The attached npm-format archive is available without npm-registry publication. This is a skills/package release; no hosted application deployment is configured.