Command Shelf

Local Claude Code mod. Install: claude plugin install command-shelf@my-claude-code-mods --scope user, then reload plugins.
Successful-command history with explicit permission-aware replay.
/shelf help
/shelf [list | show <id> | pin <id> <label> | run <id> | remove <id>]
/shelf reset
Reports are returned as command text, usable in terminal/Desktop/headless sessions. State is isolated by cwd and session ID in the plugin store; resuming the same identity restores it. Reset clears this mod's records only. Observations never block tools or execute commands twice. No automatic network calls. Explicit replay performs the selected command's normal side effects. Obvious secrets and terminal escapes are masked before storage/display; masking is best effort.
First release: see design for exact analysis limits. Git commands have a 10-second timeout. Lists cap at 200 entries and repository scans at 2,000 paths / 2 MiB; incomplete evidence is reported. Commands use explicit data entry; no model inference or automatic source editing. Actual engine tests cover command behavior and forwarding; live terminal/Desktop pixels have not been checked.
Development: npm test, CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 claude plugin test command-shelf, npx tsc -p command-shelf. Engine typechecking needs local ignored declarations. Canonical runtime sources are in scripts/workflow/; run node scripts/sync-workflow.mjs after editing.
/shelf list
/shelf pin 1 Auth regression tests
/shelf show 1
/shelf run 1
Up to 200 successful foreground commands. Commands containing recognized secrets or terminal escapes are excluded. Replay binds the original cwd and uses the Bash permission path; it never executes during listing. Other mods can observe an explicit replay. Environment values are not snapshotted. Shell cwd binding targets macOS/Linux.
Local preview before marketplace publication: claude --plugin-dir ./command-shelf from this checkout. Engine tests were run with Claude Code 2.1.302.
hooks/register.ts 91 lines1import 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 = 'command-shelf'
6const command = 'shelf'
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}
91hooks/core/runtime.ts 132 lines1import { 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}
132hooks/core/feature.ts 34 lines1import { bounded, outcome, restore, safeText, type Observation, type Port } from './runtime.ts'
2type Entry = { id: number; command: string; cwd: string; label: string }
3const HELP = '/shelf [list | show <id> | pin <id> <label> | run <id> | remove <id> | reset] — only explicit run executes through Bash permissions.'
4export class Feature {
5 data: { nextId: number; commands: Entry[] }
6 constructor(value?: unknown) { this.data = restore(value, { nextId: 1, commands: [] as Entry[] }) }
7 async observe(e: Observation, io: Port): Promise<void> {
8 if (e.tool !== 'Bash' || outcome(e.result) !== 'pass') return
9 const command = e.input.command
10 if (typeof command !== 'string' || !command.trim() || command.length > 4000 || safeText(command) !== command || safeText(io.cwd) !== io.cwd) return
11 if (this.data.commands.some(c => c.command === command && c.cwd === io.cwd)) return
12 this.data.commands = bounded([...this.data.commands, { id: this.data.nextId++, command, cwd: io.cwd, label: '' }])
13 }
14 async command(args: string, io: Port): Promise<string> {
15 const [verb, idText, ...rest] = args.trim().split(/\s+/)
16 if (verb === 'help') return HELP
17 if (!verb || verb === 'list') return ['Command Shelf — successful foreground commands, latest 200; environment is not captured.', ...this.data.commands.map(c => `#${c.id} ${c.label ? c.label + ' — ' : ''}${c.command}\n cwd: ${c.cwd}`)].join('\n')
18 const entry = this.data.commands.find(c => c.id === Number(idText))
19 if (!['show', 'pin', 'run', 'remove'].includes(verb)) return 'Usage: ' + HELP
20 if (!entry) return 'Command not found'
21 if (verb === 'pin') { const label = rest.join(' '); if (!label) return 'Usage: /shelf pin <id> <label>'; entry.label = safeText(label, 500) }
22 if (verb === 'remove') { this.data.commands = this.data.commands.filter(c => c !== entry); return 'Command removed' }
23 if (verb === 'run') {
24 if (entry.cwd !== io.cwd) return 'Refusing replay in a different project'
25 if (safeText(entry.command) !== entry.command) return 'Refusing replay of a masked command'
26 if (!io.replay) return 'Replay unavailable in this host'
27 const quote = (s: string) => "'" + s.replace(/'/g, "'\\''") + "'"
28 const result = await io.replay(`cd -- ${quote(entry.cwd)} && ${entry.command}`)
29 return outcome(result) === 'pass' ? `Command #${entry.id} completed` : `Command #${entry.id} not completed successfully; inspect Bash result`
30 }
31 return `#${entry.id} ${entry.label}\n${entry.command}\ncwd: ${entry.cwd}\nEnvironment values are not captured; review before /shelf run ${entry.id}`
32 }
33}
34hooks/core/redact.ts 31 lines1// 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