Codelore is a portable skill and agent suite for exploring a codebase. Its 11 skills follow the Agent Skills open format, so compatible harnesses can discover them at .agents/skills. It can assemble a source-cited review of a repository or answer a focused question about its architecture, execution path, dependencies, health, and setup. The CLI installs the skills and agent profiles into a local checkout. Codelore uses the active coding session's model access and repository permissions; it does not need a separate Codelore LLM API key.
Open the repository you want to inspect in a supported coding harness. From that checkout, install Codelore's skills and agent profiles with Node.js 20 or later:
npx codelore-agent-suiteYou can also run npx codelore-agent-suite init D:\path\to\target-repo from another directory. See Harness support for options and the CLI package for source and release details. Give your coding session access to the target checkout, then ask:
Use the codelore-workflow skill to review D:\path\to\target-repo at its current revision. Cover the file map, architecture, one execution trace, dependencies, health, and setup. Cite source files and state what you could not verify.
For a narrower task, use a specialist skill such as codelore-codebase-qa or codelore-dependency-audit. A result is returned in the coding session by default; ask for a file or diagram when you need one. Using Codelore has more examples and scope guidance. Skill invocation syntax varies by harness; see Harness support.
- 11 skills for repository intake, file exploration, architecture mapping, execution tracing, dependency auditing, codebase Q&A, health insights, onboarding, visual reports, review refreshes, and the full workflow.
- 11 matching agent profiles for Codex, with generated profiles for Claude Code, Cursor, Gemini CLI, GitHub Copilot, OpenCode, and Kiro. Other supported harnesses can use the portable skills; see Harness support.
- A conflict-safe CLI that installs the skills and selected profiles into an existing checkout. It skips matching files and stops before changing any files if a selected destination differs.
- Using Codelore — choose a workflow and interpret its results.
- Agent Skills conformance — frontmatter, progressive disclosure, bundled scripts, and validation.
- Harness support — discovery paths, agent profiles, and invocation examples.
- Kit authoring — canonical content rules, reference provenance, and validation.
- How the project fits together — skill/agent roles, evidence flow, and boundaries.
- Feature coverage — all 11 skills and matching agents.
- Evidence contract and health score rubric — output rules used by the suite.
See CONTRIBUTING.md for the authoring workflow and local checks. Use GitHub Issues for bugs and feature requests.
Codelore is available under the MIT License.