igmrrf/agentic

v1.4.0MIT AND Apache-2.0

Multi-language engineering standards, agent rules, and agent skills, set up in any project with one command.

Changelog

All notable changes to this project will be documented in this file.

Explicit global/project skill targets for Claude, Codex, Pi, OpenCode, Antigravity CLI/App, Gemini CLI, Cursor, Copilot, Cline, Windsurf, and Amp, with aliases and destination deduplication.
Registry-generated portable and Claude plugin manifests, a Codex marketplace, Pi package metadata, and strict validation of the packed distribution.
Harness installation lifecycle coverage, package content checks, and installation/use guides linked from a README covering the complete engineering workflow.

Project initialization now includes AGENTS.md by default and accepts shared-instruction harness targets. Setup selects the current harness instead of defaulting to Claude.
Explicit --agent all installs every harness location; the implicit default remains Claude plus shared .agents skills. Named Antigravity and Gemini installs now use their native global paths.
npm distributions exclude development context and local build/cache artifacts while retaining standards, standalone resources, and licenses.

Standalone Caveman core conversational skill, pinned to 9b68f7ed12b05e7efaff18e65c5a3fc2849fd0a4, with explicit activation, truthful status, artifact/evidence preservation, and current Apache-2.0 plus historical MIT notices. Infrastructure and alternate skills/aliases are excluded.
Complete Security Audit skill, pinned to c1c8a8c1471069fb0e188eeaff69b8e8db6564a8, with guidance/full-audit modes, all six phases, attack references, schema, validators/tests and Cloudflare MIT notice. Missing independence or OS enforcement stays an explicit runtime blocker.
Registry-generated discovery for both skills, standalone installation/package regressions, upstream validator CI checks, protected Codinary contract digests, and documented ownership, provenance and behavioral evidence. Codinary delivery rules remain unchanged; audit findings grant no remediation or release authority.

scripts/init.sh: split from one 868-line script into a 120-line entry point and focused files under scripts/lib/init/ (cli, language, files, toolchains, rules, run). The per-language rules headers moved from heredocs into templates/rules/<language>.md|.mdc. All 70 characterization scenarios are unchanged.
scripts/init.sh: remote runs now execute the downloaded ref's own init.sh, so --ref always runs that version's code (including v1.0.0 and v1.1.0).
scripts/init.sh, scripts/install-skills.sh: temporary downloads honour TMPDIR on macOS too.
/agentic:setup skill: reflects that existing files, including agent rules files, are kept unless the user agrees to --force --backup. Plugin version bumped to 1.2.0.

scripts/init.sh: existing agent rules files (CLAUDE.md, GEMINI.md, .cursor/rules/coding.mdc, .clinerules, .windsurfrules, .github/copilot-instructions.md) were silently overwritten even without --force. They are now kept like every other existing file; pass --force (and --backup for .bak copies) to regenerate them, for example after updating the standards.

tests/init/: 70 characterization scenarios for scripts/init.sh (every language, starter, flag, alias, agent selection, conflict, starter-skip and auto-detection path), run in CI under Linux bash and macOS bash 3.2. They pin the current behaviour before init.sh is split into smaller files.

/agentic:setup skill: applies the standards to the current project from inside Claude Code (dry run first, then write). Plugin version bumped to 1.1.0.
npx -y github:igmrrf/agentic: bin/agentic.js runs init and skills from npm without a clone or an npm publish.
scripts/init.sh: --ref <branch|tag|commit> for remote runs.
scripts/test-init.sh: initializer and npx launcher tests, run in CI on Linux and macOS bash 3.2.

scripts/init.sh: remote runs download the repository once as a tarball instead of one curl request per file, and --skills uses the downloaded installer.

scripts/init.sh: when piped through bash, the language prompt read from the script itself; it now reads from the terminal, or exits asking for --lang when there is none.
scripts/init.sh: failed downloads and missing standards files now stop the run instead of writing empty agent rules.

Repository renamed from igmrrf/Agentic to igmrrf/agentic; all install URLs now use the lowercase name (GitHub redirects the old one).
scripts/init.sh: Updated the scaffold script to directly embed the universal and language-specific coding standards into the generated agent files (CLAUDE.md, GEMINI.md, and .cursor/rules/coding.mdc).
scripts/init.sh: Stopped generating raw CODING.md and docs/CODING_*.md files in target repositories, preventing duplicate files and ensuring LLMs natively parse the instructions.
Agent Rules: Removed markdown references and links to the now-deprecated raw CODING.md files from the agent rule templates.

scripts/install-skills.sh: Remote (curl | bash) installs no longer install zero skills while reporting success. Log output was being captured into the skills source path. The downloaded archive is now cleaned up too.
scripts/install-skills.sh: Skills now install into .claude/skills/, where Claude Code actually reads them, as well as .agents/skills/.
scripts/install-skills.sh: --skill all works; options missing a value, empty inventories, missing target directories, and failed downloads now exit with an error instead of continuing.

Claude Code plugin marketplace (.claude-plugin/marketplace.json): /plugin marketplace add igmrrf/agentic then /plugin install agentic@agentic.
scripts/install-skills.sh: --agent claude|agents|all, --ref <branch|tag|commit>, --uninstall, and --update.
scripts/test-install-skills.sh and the skills CI workflow: installer smoke tests on Linux and macOS bash 3.2, shellcheck, marketplace validation, and a check that the marketplace lists every skill.
scripts/init.sh: Added support for generating multiple agent rule files in a single run (e.g. --claude --gemini or -a claude,gemini).
New Agents: Out-of-the-box support for Cline (.clinerules), Windsurf (.windsurfrules), and GitHub Copilot (.github/copilot-instructions.md). These files automatically embed universal and language-specific coding standards just like Claude and Gemini files.
Swift & Kotlin Standards: Added full standards, toolchain configurations, starter templates, and CI workflows for Swift 6.0+ and Kotlin 2.0+ (K2 compiler).