AGENT SKILL

prism-maintain

Maintain the Prism codebase: measure cyclomatic complexity, refactor hotspots, and enforce the anti-slop type discipline. Use when the user asks about "complexity", "refactor", "hotspot", "maintainability", "code health", "anti-slop", "lint rules", or "clean up" the Prism repo.

License 许可
MIT
最近更新
2026-10-08
Stars
★ 11
Forks
3

01 前言 Front Matter

name
prism-maintain
tags
[ "typescript", "effect-ts", "refactoring", "complexity", "lint", "maintainability" ]
author
irfndi
license
MIT
version
0.1.0
homepage
https://github.com/irfndi/prism-liquidity-agent
description
Maintain the Prism codebase: measure cyclomatic complexity, refactor hotspots, and enforce the anti-slop type discipline. Use when the user asks about "complexity", "refactor", "hotspot", "maintainability", "code health", "anti-slop", "lint rules", or "clean up" the Prism repo.
compatibility
Requires Bun 1.4+ and oxlint with the vendored anti-slop plugin.

03 SKILL.md 正文

Prism Code Maintenance

Measure complexity, refactor hotspots, keep the code human-maintainable. Methodology adapted from saurabhkumar8112/cyclomatic-complexity-skill (Apache-2.0); type discipline enforced by the vendored dmmulroy/anti-slop v0.1.2 plugin in tools/oxlint/anti-slop/.

Complexity gates (project config wins)

bun run lint enforces complexity: max 10 as an error via oxlint (type-aware — it is the authoritative measure for this Effect-TS codebase; line-counters like lizard mis-parse Effect.gen chains and report phantom CCN 50+ on 15-line functions, so never act on lizard numbers here).

Bands: 1–6 fine · 7–10 watch (refactor only if touching the function anyway) · 11+ must split before merge (lint fails).

Watch-band probe (never commit this config):

sed '"'"'s/"max": 10/"max": 6/'"'"' .oxlintrc.json > .oxlint-cc6.json
bunx oxlint --config .oxlint-cc6.json engine ops bench cli
rm .oxlint-cc6.json

Workflow (mandatory)

  1. Measure the touched functions, rank by CCN descending.
  2. Report hotspots with numbers BEFORE touching anything.
  3. Refactor worst first, one function at a time.
  4. Re-measure; close with the report table below.
  5. Verify: bun run lint, bun run format, full bun run test green with ZERO test edits (unchanged tests passing is the behavior proof).
## Complexity report
| Function | Before | After |
|----------|--------|-------|
| runPoolEntryGates | 11 | 9 |
Extracted: isSafetyBinsSkippable, snapshotPriceDrift
Behavior verified: full suite green, no test edits

Refactor tactics (preference order)

  1. Guard clauses — invert, return early, kill nesting.
  2. Extract function — name for WHAT, not how.
  3. Lookup table / map over if-else chains.
  4. Named predicates — isPositiveFiniteApr(x) beats boolean soup.
  5. Lookup before polymorphism: Effect code dispatches via Layers and tagged unions, not class hierarchies.
  6. continue over nested if in loops.

Hard rules: preserve behavior · never game the metric (a dense one-liner hiding 6 branches is worse than the honest chain) · never break exported signatures without asking · one responsibility per function.

When NOT to split

  • Numbered risk-gate chains (engine/risk-service.ts): sequential early-return blocks are the documented convention (AGENTS.md) — splitting scatters the gate narrative.
  • Single decision-loop Effect.gen blocks (engine/program.ts): splitting obscures per-cycle sequencing; extract pure predicates instead.
  • Agent-overlay handlers and network retry logic in untouched domains.

Anti-slop plugin (vendored v0.1.2 parity)

Source: tools/oxlint/anti-slop/ (.ts sources + index.mjs bundles + upstream *.test.ts contracts, which need oxlint/plugins-dev to execute). Rebuild bundles after any source change — never hand-edit .mjs:

bun build tools/oxlint/anti-slop/index.ts --outfile tools/oxlint/anti-slop/index.mjs
bun build tools/oxlint/anti-slop/effect/index.ts --outfile tools/oxlint/anti-slop/effect/index.mjs

Rules that most often bite: no-known-value-widening (open-dict annotations — prefer finite type contracts, satisfies, accumulators, or Map), require-safety-comment-for-type-assertion (every as needs a SAFETY: justification with non-empty text), no-unknown-parameters / no-unknown-returns (parse at the boundary, return named domain types). Never suppress or weaken a rule — fix the code.

04 同仓库更多 Skills