Changelog

All notable changes to this project are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Real LTF trace emission: the generated verification hook now appends spec-shaped .loop/trace.ltf.jsonl entries (verify/terminate phase events and a loop_summary on success) instead of only generating a config file that described intended tracing.
Multi-agent scaffolding: kit init can now generate configs for multiple agents in one run via the wizard's "All (multi-agent)" option or --agent <agents> (comma-separated or all). A .loop/kit.json manifest records which agents were selected.
kit score gives proportional credit for multi-agent setups based on the recorded manifest, falling back to the original any-one-file check for pre-existing projects.
Cursor now falls back to a template's claude-code instructions when no cursor-specific instructions are defined (no shipped template currently defines them).
A one-time warning when running on Windows without bash on PATH, since the generated verification hook is a bash script (documented prerequisite: Git Bash or WSL).
CI now includes a Windows smoke job in addition to the existing Linux matrix.

Critical: a verify command containing a literal exit call (a common shell idiom) would terminate the entire generated hook script before capturing the exit code or emitting any trace — never surfaced because no prior test exercised this pattern. Fixed by running the verify command in a subshell.
src/cli/status.ts's trace reader modeled tokens as a flat number instead of the real LTF nested {input, output, cached, cache_write} shape, and had no concept of loop_summary records; both are now handled correctly, with parse-failure counts surfaced instead of silently dropped.
.loop/ltf.config.yaml's field list omitted ltf_version/loop_id/phase — the four fields every real LTF event requires — while claiming to describe the trace schema. Corrected to state plainly what's captured (verify-phase events only; no cost/token data, since the hook has no visibility into agent model calls).

Initial release.

kit init — interactive Clack-based wizard, --yes non-interactive mode, and --template <name> for scaffolding from a built-in template.
Auto-detection of project stack (TypeScript/JavaScript/Python/Go/Rust), test runner (vitest/jest/mocha/pytest/go-test/cargo-test), linter (ESLint/Ruff), and installed agent CLIs.
Generators for CLAUDE.md, AGENTS.md, GEMINI.md, .cursorrules, a bash verification hook, .loop/goal.md, .loop/budget.yaml, and .loop/ltf.config.yaml.
kit templates — browse and search 6 built-in templates (fix-types, fix-lint, add-tests, migrate-api, dependency-update, security-audit).
kit score — loop readiness score (0-100) across 10 criteria.
kit status — reads .loop/state.md, .loop/budget.yaml, and LTF traces.