assumption-tracker

1import type { EngineInterface, Register } from 'claude-code'
2import { object, revision, safeText, safeData, reportText, type Observation, type Port } from './core/runtime.ts'
3export interface Model { data: object; command(args: string, io: Port): Promise<string>; observe?(event: Observation, io: Port): Promise<void> }
4import { Feature } from './core/feature.ts'
5const name = 'assumption-tracker'
6const command = 'assume'
7type AdapterState = { queue: Promise<unknown>; replaying: boolean }
8const state: AdapterState = { queue: Promise.resolve(), replaying: false }
9const observesTools = 'observe' in Feature.prototype
10const serial = <T>(fn: () => Promise<T>): Promise<T> => {
11  const work = state.queue.then(fn); state.queue = work.catch(() => {}); return work
12}
13const port = async ($: EngineInterface): Promise<Port> => {
14  const cwd = await $.session.cwd()
15  const io: Port = {
16    cwd,
17    read: async path => await $.fs.read(path),
18    bytes: async path => (await $.fs.read(path, { as: 'bytes' })).base64,
19    stat: path => $.fs.stat(path),
20    list: async path => (await $.fs.list(path)).map(entry => entry.name),
21    run: argv => $.process.run(argv, { cwd, timeoutMs: 10_000 }),
22    env: async key => {
23      if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(key)) throw new Error('invalid environment variable name')
24      const r = await $.process.run(['node', '-e', 'process.stdout.write(process.env[process.argv[1]] ? "set" : "")', key], { cwd, timeoutMs: 10_000 })
25      if (r.exitCode !== 0) throw new Error('environment presence unavailable')
26      return r.stdout === 'set' ? 'set' : undefined
27    },
28    revision: () => revision(io),
29    replay: async text => {
30      state.replaying = true
31      try { return await $.tool.call({ tool: 'Bash', command: text, description: 'Replay explicitly selected Command Shelf entry' }) }
32      finally { state.replaying = false }
33    },
34  }
35  return io
36}
37const identity = ($: EngineInterface) => Promise.all([$.session.cwd(), $.session.id()])
38const transact = async ($: EngineInterface, fn: (model: Model, io: Port) => Promise<string | void>, expected: string[], forceSave = false): Promise<string> => serial(async () => {
39  const [io, id] = await Promise.all([port($), $.session.id()])
40  if (io.cwd !== expected[0] || id !== expected[1]) return 'Session changed; operation not applied'
41  const key = `${name}:v1:${JSON.stringify([io.cwd, id])}`
42  const raw = await $.store.get(key)
43  const loaded = await identity($)
44  if (loaded[0] !== expected[0] || loaded[1] !== expected[1]) return 'Session changed; operation not applied'
45  const model = new Feature(raw)
46  const previous = JSON.stringify(safeData(model.data))
47  const result = await fn(model, io)
48  // No raw event payloads reach storage. Keep the serialized state bounded.
49  const serialized = JSON.stringify(safeData(model.data))
50  if (forceSave || serialized !== previous) {
51    if (serialized.length > 256 * 1024) throw new Error('state limit exceeded; not saved')
52    try { await $.store.set(key, JSON.parse(serialized)) }
53    catch { throw new Error('state not saved; retry after storage recovers') }
54  }
55  return reportText(result ?? '')
56})
57export const register: Register = on => {
58  on('session.start', async ($, e, next) => {
59    try { await $.command.register({ name: command, description: `${name}: /${command} help` }) } catch { /* Startup still forwards. */ }
60    return next(e)
61  })
62  on('command.run', { command }, async ($, e) => {
63    try {
64      const expected = await identity($)
65      return { text: await transact($, async (model, io) => {
66        if ((e.args ?? '').trim() === 'reset') { model.data = new Feature().data; return `${name}: records reset` }
67        return model.command(e.args ?? '', io)
68    }, expected, (e.args ?? '').trim() === 'reset') } } catch (error) { return { text: `${name}: ${safeText(error instanceof Error ? error.message : error)}` } }
69  })
70  on('tool.call', async ($, e, next) => {
71    const ownReplay = state.replaying && (next.origin?.plugin === name || next.origin?.plugin.startsWith(name + '@'))
72    if (!observesTools || ownReplay) return next(e)
73    const expected = await identity($).catch(() => undefined)
74    let before: string | undefined
75    if (['verify-board', 'handoff-pack'].includes(name) && e.tool === 'Bash') {
76      try { before = await (await port($)).revision() } catch { /* Unknown evidence. */ }
77    }
78    // next runs outside observation catch: an observer never retries a tool or hides its exception.
79    const result = await next(e)
80    try {
81      const now = await Promise.all([$.session.cwd(), $.session.id()])
82      if (!expected || expected[0] !== now[0] || expected[1] !== now[1]) return result
83      await transact($, async (model, io) => {
84        const after = before !== undefined ? await io.revision() : undefined
85        await model.observe?.({ tool: e.tool, input: object(e), result, before, after }, io)
86      }, expected)
87    } catch { try { $.ui.toast(`${name}: observation not saved`) } catch { /* Noninteractive host. */ } }
88    return result
89  })
90}
91

1import { redactSecrets } from './redact.ts'
2export type ProcessResult = { exitCode: number; stdout: string; stderr: string; isStdoutTruncated?: boolean; isStderrTruncated?: boolean }
3export type Port = {
4  cwd: string
5  read(path: string): Promise<string>
6  bytes?(path: string): Promise<string>
7  stat?(path: string): Promise<{ isLink: boolean }>
8  list?(path: string): Promise<string[]>
9  run(argv: readonly string[]): Promise<ProcessResult>
10  env(name: string): Promise<string | undefined>
11  revision(): Promise<string | undefined>
12  replay?(command: string): Promise<unknown>
13}
14export type Observation = { tool: string; input: Record<string, unknown>; result: unknown; before?: string; after?: string }
15export function object(v: unknown): Record<string, unknown> { return v !== null && typeof v === 'object' && !Array.isArray(v) ? v as Record<string, unknown> : {} }
16export function safeText(v: unknown, limit = 4000): string {
17  const text = String(v ?? '').replace(/\x1b\][^\x07]*(?:\x07|\x1b\\)/g, '').replace(/\x1b\[[0-?]*[ -/]*[@-~]/g, '').replace(/[\x00-\x08\x0b-\x1f\x7f-\x9f]/g, '')
18  // Some masking patterns backtrack on very long words. Omit an oversized line whole:
19  // cutting it first could expose an unterminated quoted secret.
20  const lines = text.split('\n')
21  // Dropping a credential's opening line could expose its quoted continuation.
22  if (lines.some(line => line.length > 8192 && /["'\\]|password|passwd|pwd|token|key|secret|authorization|credential|cookie/i.test(line))) return '[OMITTED: oversized sensitive value]'
23  return redactSecrets(lines.map(line => line.length > 8192 ? '[OMITTED: oversized line]' : line).join('\n')).slice(0, limit)
24}
25export function hash(text: string): string {
26  return [0, 0x9747b28c].map(seed => {
27    let h = (0x811c9dc5 ^ seed) >>> 0
28    for (let i = 0; i < text.length; i++) h = Math.imul(h ^ text.charCodeAt(i), 0x01000193)
29    return (h >>> 0).toString(16).padStart(8, '0')
30  }).join('')
31}
32export function relativePath(root: string, path: string): string | undefined {
33  const normalize = (p: string) => {
34    const parts: string[] = []
35    for (const s of p.replace(/\\/g, '/').split('/')) {
36      if (!s || s === '.') continue
37      if (s === '..') { if (!parts.length) return undefined; parts.pop() } else parts.push(s)
38    }
39    return parts.join('/')
40  }
41  const base = normalize(root); if (base === undefined) return undefined
42  const full = normalize(path.startsWith('/') ? path : `${root}/${path}`)
43  if (full === undefined) return undefined
44  if (base === '') return root.startsWith('/') ? full : undefined
45  if (!(full === base || full.startsWith(base + '/'))) return undefined
46  return full.slice(base.length).replace(/^\//, '')
47}
48export function changedPaths(status: string): string[] {
49  const entries = status.split('\0'); const paths: string[] = []
50  for (let i = 0; i < entries.length; i++) {
51    const line = entries[i]!; if (line.length < 4) continue
52    paths.push(line.slice(3))
53    if (/[RC]/.test(line.slice(0, 2)) && entries[i + 1]) paths.push(entries[++i]!)
54  }
55  return [...new Set(paths)]
56}
57export async function git(io: Port, args: string[]): Promise<string> {
58  const r = await io.run(['git', '-C', io.cwd, ...args])
59  if (r.exitCode !== 0 || r.isStdoutTruncated || r.isStderrTruncated) throw new Error('Git evidence unavailable or incomplete')
60  if (r.stdout.length > 2 * 1024 * 1024) throw new Error('Git evidence exceeds 2 MiB limit')
61  return r.stdout
62}
63export async function statusPaths(io: Port): Promise<string[]> { return changedPaths(await git(io, ['status', '--porcelain=v1', '-z', '--untracked-files=all'])) }
64export async function revision(io: Port): Promise<string | undefined> {
65  try {
66    const root = (await git(io, ['rev-parse', '--show-toplevel'])).replace(/\n$/, '')
67    if (!root.startsWith('/')) return undefined
68    const repo = { ...io, cwd: root }
69    const head = await git(repo, ['rev-parse', '--verify', 'HEAD'])
70    const diff = await git(repo, ['diff', '--no-ext-diff', '--no-textconv', '--binary', '--ignore-submodules=none', 'HEAD', '--'])
71    const staged = await git(repo, ['diff', '--no-ext-diff', '--no-textconv', '--binary', '--ignore-submodules=none', '--cached', 'HEAD', '--'])
72    if (/^\+Subproject commit [a-f0-9]+-dirty$/m.test(diff) || /^\+Subproject commit [a-f0-9]+-dirty$/m.test(staged)) return undefined
73    const status = await git(repo, ['status', '--porcelain=v1', '-z', '--untracked-files=all', '--ignore-submodules=none'])
74    const paths = (await git(repo, ['ls-files', '--others', '--exclude-standard', '-z'])).split('\0').filter(Boolean).sort()
75    if (paths.length > 200) return undefined
76    const bodies: string[] = [head, diff, staged, status]; let size = diff.length + staged.length + status.length
77    for (const path of paths) {
78      if (relativePath(root, path) !== path || (io.stat && (await io.stat(`${root}/${path}`)).isLink)) return undefined
79      const body = await (io.bytes ?? io.read)(`${root}/${path}`); size += body.length
80      if (size > 2 * 1024 * 1024) return undefined
81      bodies.push(path, body)
82    }
83    return hash(JSON.stringify(bodies))
84  } catch { return undefined }
85}
86export function outcome(result: unknown): 'pass' | 'fail' | 'unknown' {
87  const a = object(result), r = object(a.result)
88  if (a.deny !== undefined) return 'unknown'
89  for (const v of [a, r]) {
90    if (['interrupted', 'cancelled', 'canceled', 'timedOut', 'timed_out', 'truncated', 'output_truncated', 'isStdoutTruncated', 'isStderrTruncated', 'backgroundedByUser', 'backgroundedByTurnAbort', 'backgroundedToDeliverMessage'].some(k => v[k] === true)) return 'unknown'
91    if (['task_id', 'taskId', 'backgroundTaskId', 'background_task_id', 'timedOutAfterMs', 'rawOutputPath', 'persistedOutputPath'].some(k => v[k] !== undefined && v[k] !== null && v[k] !== '')) return 'unknown'
92  }
93  if (a.isError === true) return 'fail'
94  const text = [a.text, r.stdout, r.stderr].filter((v): v is string => typeof v === 'string').join('\n').replace(/\x1b\[[0-?]*[ -/]*[@-~]/g, '')
95  if (text.length > 65_536) return 'unknown'
96  if (/^# fail [1-9]\d*\s*$/m.test(text) || /^\s*Tests(?::|\s+)[^\n]*\b[1-9]\d* failed\b/m.test(text) || /^=+[^\n]*\b[1-9]\d* (?:failed|errors?)\b/m.test(text) || /^test result: FAILED\./m.test(text) || /^\s*--- FAIL: /m.test(text)) return 'fail'
97  const code = 'exitCode' in r ? r.exitCode : r.exit_code
98  if ('exitCode' in r || 'exit_code' in r) return typeof code === 'number' && Number.isInteger(code) ? code === 0 ? 'pass' : 'fail' : 'unknown'
99  // Builtin Bash exposes tool failure via isError and foreground completion via interrupted/streams, not exitCode.
100  return typeof r.stdout === 'string' && typeof r.stderr === 'string' && r.interrupted === false ? 'pass' : 'unknown'
101}
102export function isCheck(command: string): boolean {
103  if (/^(?:npx\s+)?vitest\b.*\slist(?:\s|$)/.test(command.trim())) return false
104  // Known introspection-only options do not run the check itself.
105  if (/(?:^|\s)(?:--help|-h|--version|-V|--collect-only|--co|--listTests|--list-tests|--showConfig|--show-config|--list|-list)(?:\s|=|$)/.test(command)) return false
106  // Conservative whole-command allowlist; compound shell programs are not proof of a check.
107  return /^(?:(?:npm|pnpm|yarn) (?:run )?(?:test(?::[\w-]+)?|typecheck|lint)|npx (?:tsc|vitest|jest|eslint)|(?:pytest|vitest|jest|tsc|eslint)|cargo (?:test|check|clippy)|go test)\b[^;&|<>\n]*$/.test(command.trim())
108}
109export function restore<T extends object>(value: unknown, defaults: T): T {
110  const v = object(value); const out = { ...defaults } as Record<string, unknown>
111  for (const [k, fallback] of Object.entries(defaults)) {
112    const item = v[k]
113    if (Array.isArray(fallback) ? Array.isArray(item) : typeof item === typeof fallback && item !== null) out[k] = item
114  }
115  return out as T
116}
117export const entries = (args: string): string[] => args.split('|').map(s => s.trim())
118export function bounded<T>(items: T[]): T[] { return items.slice(-200) }
119// Clean string leaves before JSON serialization, never regex-rewrite serialized JSON syntax.
120export function safeData(value: unknown): unknown {
121  if (typeof value === 'string') return safeText(value)
122  if (Array.isArray(value)) return value.map(safeData)
123  if (value !== null && typeof value === 'object') return Object.fromEntries(Object.entries(value).map(([k, v]) => [k, safeData(v)]))
124  return value
125}
126
127export function reportText(value: unknown): string {
128  const limit = 32 * 1024, notice = '\nINCOMPLETE: report truncated; filter or inspect smaller groups.'
129  const text = safeText(value, limit + 1)
130  return text.length > limit || String(value ?? '').length > limit ? text.slice(0, limit - notice.length) + notice : text
131}
132

1import { bounded, entries, restore, safeText, type Port } from './runtime.ts'
2type Status = 'UNVERIFIED' | 'CONFIRMED' | 'REFUTED'
3type Assumption = { id: number; claim: string; reference: string; status: Status; evidence: string; history: { status: Status; evidence: string }[] }
4const HELP = '/assume [add <claim> | <reference optional>] [list | show <id> | confirm <id> | <evidence> | refute <id> | <evidence> | reopen <id> | reset]'
5export class Feature {
6  data: { nextId: number; assumptions: Assumption[] }
7  constructor(value?: unknown) { this.data = restore(value, { nextId: 1, assumptions: [] as Assumption[] }) }
8  async command(args: string, _io: Port): Promise<string> {
9    const [verb, ...rest] = args.trim().split(/\s+/); const text = rest.join(' ')
10    if (verb === 'help') return HELP
11    if (verb === 'add') {
12      const [claim, reference = ''] = entries(text)
13      if (!claim) return 'Usage: ' + HELP
14      this.data.assumptions = bounded([...this.data.assumptions, { id: this.data.nextId++, claim: safeText(claim, 2000), reference: safeText(reference, 500), status: 'UNVERIFIED', evidence: '', history: [] }])
15    } else if (['confirm', 'refute', 'reopen', 'show'].includes(verb ?? '')) {
16      const [id, evidence = ''] = entries(text); const item = this.data.assumptions.find(a => a.id === Number(id))
17      if (!item) return 'Assumption not found'
18      if (verb === 'show') return `#${item.id} ${item.status} ${item.claim}\n${item.reference}\n${item.history.map(h => `${h.status}: ${h.evidence}`).join('\n')}`
19      if (verb !== 'reopen' && !evidence) return 'Resolution evidence required: confirm/refute <id> | <evidence>'
20      item.status = verb === 'confirm' ? 'CONFIRMED' : verb === 'refute' ? 'REFUTED' : 'UNVERIFIED'
21      item.evidence = verb === 'reopen' ? '' : safeText(evidence, 2000)
22      item.history = [...item.history, { status: item.status, evidence: item.evidence || '(reopened)' }].slice(-20)
23    } else if (verb && verb !== 'list') return 'Usage: ' + HELP
24    return ['Assumption Tracker — explicit claims, latest 200; last 20 transitions per claim', ...this.data.assumptions.map(a => `#${a.id} ${a.status} ${a.claim}\n  Ref: ${a.reference || '(none)'}\n  Evidence: ${a.evidence || '(unverified)'}`)].join('\n')
25  }
26}
27

1// Canonical source; run node scripts/sync-redactors.mjs after editing.
2// Best-effort removal of obvious secrets from command lines before they are exported.
3const VALUE = String.raw`("(?:\\[\s\S]|[^"\\])*"|'[^']*'|(?:\\[\s\S]|[^\s'"\\])+)`
4const HEADER = String.raw`(?:authorization|[a-z0-9][a-z0-9-]*-(?:key|token|secret))`
5// Only match user:password arguments, leaving username-only options (e.g. mysql -u) alone.
6const CREDENTIAL = String.raw`("(?:\\[\s\S]|[^":\\])*:(?:\\[\s\S]|[^"\\])*"|'[^':]*:[^']*'|[^\s'":]*:[^\s'"]*)`
7const RULES: [RegExp, string][] = [
8  // Shell-quoted headers must be handled before key/value rules: an escaped quote is
9  // part of the secret, while the opposite quote is ordinary data inside this shell word.
10  [new RegExp(String.raw`("` + HEADER + String.raw`\s*:\s*)(?:\\[\s\S]|[^"\\])*"`, 'gi'), '$1[REDACTED]"'],
11  [new RegExp(String.raw`('` + HEADER + String.raw`\s*:\s*)[^']*'`, 'gi'), "$1[REDACTED]'"],
12  [new RegExp(String.raw`(\b` + HEADER + String.raw`\s*:\s*)(?:\\[\s\S]|[^'"\r\n\\])+`, 'gi'), '$1[REDACTED]'],
13  [new RegExp(String.raw`((?:["']?)(?:password|passwd|pwd|token|(?:access|refresh|id|session|auth)[_-]?token|(?:api|access|secret|private)[_-]?key|client[_-]?secret|secret|authorization|credentials?|cookies?)(?:["']?)\s*:\s*)` + VALUE, 'gi'), '$1[REDACTED]'],
14  [/(\bbearer\s+)[A-Za-z0-9._~+/=-]+/gi, '$1[REDACTED]'],
15  [new RegExp(String.raw`(--(?:password|token|api-key|secret)(?:=|\s+))` + VALUE, 'gi'), '$1[REDACTED]'],
16  [/([a-z][a-z0-9+.-]*:\/\/[^\s/:@'"]*:)[^\s/@'"]*@/gi, '$1[REDACTED]@'],
17  [new RegExp(String.raw`(\s(?:--user(?:=|\s+)|-u\s*))` + CREDENTIAL, 'g'), '$1[REDACTED]'],
18  [/(\bmysql\w*\b[^\n;&|]*?\s-p)(?=\S)("[^"]*"|'[^']*'|\S+)/g, '$1[REDACTED]'],
19  [new RegExp(String.raw`\b([A-Za-z0-9_]*(?:KEY|TOKEN|SECRET|PASSWORD)[A-Za-z0-9_]*)=` + VALUE, 'gi'), '$1=[REDACTED]'],
20]
21
22export function redactSecrets(text: string): string {
23  return RULES.reduce((out, [pattern, replacement]) => out.replace(pattern, replacement), text)
24}
25
26// Match complete key names, including common camelCase/underscore/hyphen spellings.
27// Do not match ordinary names such as token_count or keyboard.
28export function isSecretKey(key: string): boolean {
29  return /^(?:password|passwd|pwd|authorization|proxyauthorization|apikey|accesskey|secretkey|privatekey|clientsecret|secret|token|accesstoken|refreshtoken|idtoken|sessiontoken|authtoken|credential|credentials|cookie|setcookie)$/i.test(key.replace(/[_\s-]/g, '')) || /^[a-z0-9][a-z0-9-]*-(?:key|token|secret)$/i.test(key)
30}
31