verify-before-done

1// verify-before-done: it watches each turn for code edits (Edit, Write, MultiEdit, NotebookEdit on a code file) and for
2// test, check or build commands (Bash). When a turn ends having changed code without running any, the person gets a
3// toast at once, and the model gets one reminder line with the person's next prompt, asking it to say what was and was
4// not verified. It never blocks and never runs anything. FAIL OPEN: any error leaves the turn and the prompt unchanged.
5import { atom, read, update } from 'claude-code'
6import type { EngineInterface, Register } from 'claude-code'
7
8import { MISSED_TOAST, NOTE, isCodeFile, isEditTool, isTestCommand } from './lib'
9
10type Dollar = EngineInterface
11
12const COMMAND = 'verify-before-done'
13
14const isOff = atom({ plugin: 'verify-before-done', key: 'isOff' } as const, false)
15const unverified = atom({ plugin: 'verify-before-done', key: 'unverified' } as const, 0)
16
17// This turn's facts and the reminder waiting for the next prompt (the module's own variables: they restart on a reload).
18let edited = false
19let tested = false
20let owed = false
21
22async function run($: Dollar, args: string): Promise<string> {
23  const word = args.trim().toLowerCase()
24  if (word === 'on' || word === 'off') {
25    await $.store.set('off', word === 'off')
26    await update($, isOff, () => word === 'off')
27    return `verify-before-done · ${word}`
28  }
29  if (word === '' || word === 'status') {
30    const n = await read($, unverified)
31    return `verify-before-done · ${(await read($, isOff)) ? 'off' : 'on'} · ${n} turn${n === 1 ? '' : 's'} changed code with no test run this session`
32  }
33  return 'verify-before-done · use /verify-before-done [on|off|status]; nothing changed'
34}
35
36export const register: Register = on => {
37  on('session.start', async ($, e, next) => {
38    edited = tested = owed = false
39    try {
40      const off = (await $.store.get('off')) === true
41      await update($, isOff, () => off)
42    } catch {
43      // reads as on
44    }
45    try {
46      await $.command.register({ name: COMMAND, description: 'Flags a turn that changed code and ran no test: /verify-before-done [on|off|status]', argumentHint: '[on|off|status]', immediate: true })
47    } catch {
48      // fail-open: the mod still works without its command
49    }
50    return next(e)
51  })
52
53  on('command.run', { command: COMMAND }, async ($, e) => ({ text: await run($, e.args) })).catch(() => ({ text: 'verify-before-done · that did not work, nothing changed' }))
54
55  on('turn.start', async (_$, e, next) => {
56    if (e.agentId === undefined) edited = tested = false
57    return next(e)
58  }).catch((_$, e, next) => next(e))
59
60  // Watches only: the call runs first and its result goes back unchanged.
61  on('tool.call', async ($, e, next) => {
62    const ran = await next(e)
63    if (ran.deny !== undefined || ran.isError === true || (await read($, isOff))) return ran
64    const input = e as unknown as Record<string, unknown>
65    const tool = String(e.tool)
66    if (tool === 'Bash' && typeof input.command === 'string' && isTestCommand(input.command)) tested = true
67    const target = typeof input.file_path === 'string' ? input.file_path : typeof input.notebook_path === 'string' ? input.notebook_path : ''
68    if (isEditTool(tool) && isCodeFile(target)) edited = true
69    return ran
70  }).catch(($, e, next) => next(e))
71
72  on('turn.complete', async ($, e, next) => {
73    const out = await next(e)
74    if (e.agentId !== undefined || e.reason !== 'answer' || (await read($, isOff))) return out
75    if (edited && !tested) {
76      owed = true
77      await update($, unverified, n => n + 1)
78      $.ui.toast(MISSED_TOAST, { timeoutMs: 8000 })
79    }
80    edited = tested = false
81    return out
82  }).catch(($, e, next) => next(e))
83
84  on('prompt.submit', async ($, e, next) => {
85    if (!owed || (await read($, isOff))) return next(e)
86    owed = false
87    return next({ ...e, context: [...(e.context ?? []), NOTE] })
88  }).catch(($, e, next) => next(e))
89}
90

1// Pure helpers for verify-before-done: what counts as code, and what counts as a test, check or build run.
2
3const EDIT_TOOLS = new Set(['Edit', 'Write', 'MultiEdit', 'NotebookEdit'])
4const CODE = /\.(r|rs|py|jl|cpp|c|h|ts|tsx|js|jsx|go|sh|sql)$/i
5
6export const isEditTool = (tool: string): boolean => EDIT_TOOLS.has(tool)
7export const isCodeFile = (path: string): boolean => CODE.test(path.replace(/\\/g, '/'))
8
9/** A command that runs tests, a check or a build of the project. */
10export const isTestCommand = (cmd: string): boolean =>
11  /\b(devtools::(test|check)|testthat|R\s+CMD\s+check|rcmdcheck|cargo\s+(test|nextest|clippy|check|build)|pytest|py\.test|npm\s+(run\s+)?(test|build)|pnpm\s+(run\s+)?(test|build)|vitest|jest|go\s+test|make\s+(test|check)|bun\s+test|claude\s+plugin\s+(test|validate)|quarto\s+render|tsc\b)/i.test(cmd)
12
13export const MISSED_TOAST = 'verify-before-done: code changed and no test or check ran this turn'
14export const NOTE =
15  'verify-before-done: your last turn changed code and ran no test, check or build. In your next reply say what was and was not verified; if you said it was done, run the tests now or retract it.'
16

1declare module 'claude-code' {
2  interface PluginState {
3    'verify-before-done': {
4      /** True while switched off (it ships on). */
5      isOff: boolean
6      /** Turns this session that changed code and ran no test, check or build. */
7      unverified: number
8    }
9  }
10}
11