Using the Framework
Initialize the framework into a repository, then drive the lifecycle with slash commands: onboard, interrogate, spec, budget, plan, implement, verify, review.
Initialize
Clone the kcc-agentic-framework repository (or add it to an existing project) and run the framework initializer for your harness. Initialization regenerates the harness-specific agents and skills from the neutral capability source.
# PowerShell powershell -ExecutionPolicy Bypass -File .KCC\tools\framework-init.ps1 -Harness claude # bash bash .KCC/tools/framework-init.sh claude
Supported harnesses include Claude Code, Codex, OpenCode, a generic runner, and an Ollama-compatible local-model target. Each produces its own generated files (for Claude: `.claude/agents/`, `.claude/skills/`, and a root `CLAUDE.md` if none exists).
The lifecycle
Work flows through an explicit lifecycle, with a human gate wherever confidence drops or budget approval is needed. Meta-agents wrap each step: the Butler gathers context before, and remembers learnings after; the Token Guard estimates cost before expensive steps.
interrogate -> create -> token-budget -> plan -> token-budget -> implement -> test -> review
The slash commands
Each lifecycle phase is a slash-command skill that delegates to the right agent:
| Command | What it does |
|---|---|
| /solution-onboard | Build a read-only baseline of an existing codebase before specs target it |
| /idea-interrogator | Turn a raw idea into a spec-ready briefing, invoking specialist interrogators |
| /spec-create | Create an epic spec folder (stories, enablers, parallelization) after interrogation |
| /token-estimate | Estimate token cost and gate it for human approval before expensive work |
| /spec-plan | Produce a dependency-waved plan plus an atomic test-case enumeration |
| /spec-implement | Implement an approved plan, writing source and tests |
| /spec-test + /spec-review | Verify the implementation against every acceptance criterion |
Greenfield vs existing code
For a new idea, start with `/idea-interrogator`. For an existing codebase, start with `/solution-onboard` so future specs reference a real solution baseline. The `auto` skill chains the whole lifecycle but stops at every human gate unless an explicit, budget-approved AutoPolicy covers it - autonomy is earned through contracts and evidence, exactly as the core thesis requires.
The framework never writes application source before interrogation, spec creation, planning, and the token-budget gate complete. Governance precedes execution; see What KCC Is Not for the boundaries it deliberately keeps.