cadre

name

cadre

description

Context-driven development methodology for organized, spec-first coding. Use when: - Project has a `cadre/` directory - User mentions specs, plans, tracks, or context-driven development - Files like `cadre/tracks.json`, `cadre/product.json`, or `cadre/workflow.json` exist - User asks about project status, implementation progress, or track management - User wants to organize development work with TDD practices - User asks for a `cadre-*` workflow (setup, newtrack, implement, debug, status, revert, validate, flag, revise, review, ship, land, archive, release, handoff, refresh, formula, artifacts) - User mentions documentation is outdated or wants to sync context with codebase changes - Project is a polyrepo control repo (`cadre/repos.json` with mode "polyrepo") spanning git-submodule product repos Interoperable across Claude Code, OpenAI Codex, Copilot, and Antigravity. Uses native Cadre state for persistent task memory across sessions.

Cadre Skill Shim

Cadre MCP is required for every Cadre workflow. Before acting, verify the MCP runtime with cadre_project using {"action":"ping"}. If Cadre MCP tools or resources are unavailable, halt and ask the user to install, enable, or restart the Cadre plugin.

Load cadre://skill-contract for the authoritative cadre.skill.v1 contract. Use cadre://workflow-protocols to discover workflow protocol resources, then load cadre://workflow-protocol?workflow=<name> for the active workflow. References and template inventory are also MCP-served through Cadre resources.

For staged approvals, stay on Cadre's packet-led review path: use the returned approval.current_stage and current review output first. Target mode may have manifest_path:null, mutates_worktree:true, and files[].target_path / review_path; bundle mode uses the manifest and temp review paths. Ask for explicit user approval before sending approval fields. Avoid extra inspection unless packet evidence cannot answer the current stage question.