Backlog — captured from a design discussion on where third-party code-quality tools fit in kit's model. No commitment to build yet; this records the decision + the gating question.
kit's model (the ruler)
kit is the deterministic gate that runs your scanners and folds their results into one report. Plugins = adapters that ingest a tool's findings. Zero LLM calls, local-first.
Decisions
- CodeScene → thin plugin, gated on API signal. The MCP server was tried and "didn't give much" — expected, since MCP is a conversational surface, the wrong shape for a gate. What kit wants is the CodeScene REST API's code-health score / hotspot delta as a crisp threshold-able pass/fail, folded in exactly like the
snyk/wiz plugins. Worth a thin plugin only if the API exposes a tractable value. Otherwise skip — don't build an adapter for noise.
- Open question (decide before building): what do CodeScene's REST endpoints actually expose (code-health, hotspots, per-commit deltas)? Verify before committing to a plugin.
- CodeRabbit → out of scope. Do not absorb. It's an LLM-based PR-review agent living in the GitHub/PR layer — contrary to kit's "zero LLM calls, one deterministic verdict" identity. Complementary, not a component: CodeRabbit reviews the PR, kit gates the environment/supply-chain. Wrapping it would break the identity.
- dev-standard → home for the policy. Clean layering:
- plugin = data source ("fetch the code-health number").
- dev-standard = the standard/threshold ("code health ≥ X or fail; hotspots must not grow"). Subjective team policy belongs in dev-standard, not hard-coded into a scanner plugin.
Net
CodeScene as a thin plugin (data source) if its REST API delivers a usable signal, with the threshold expressed in dev-standard (policy). CodeRabbit stays outside kit.
🤖 Generated with Claude Code
https://claude.ai/code/session_015ERHw6bVUz39sAuoQZ1iyg
Backlog — captured from a design discussion on where third-party code-quality tools fit in kit's model. No commitment to build yet; this records the decision + the gating question.
kit's model (the ruler)
kit is the deterministic gate that runs your scanners and folds their results into one report. Plugins = adapters that ingest a tool's findings. Zero LLM calls, local-first.
Decisions
snyk/wizplugins. Worth a thin plugin only if the API exposes a tractable value. Otherwise skip — don't build an adapter for noise.Net
CodeScene as a thin plugin (data source) if its REST API delivers a usable signal, with the threshold expressed in dev-standard (policy). CodeRabbit stays outside kit.
🤖 Generated with Claude Code
https://claude.ai/code/session_015ERHw6bVUz39sAuoQZ1iyg