switchroom 0.18.12 → 0.18.14
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/agent-scheduler/index.js +57 -9
- package/dist/auth-broker/index.js +174 -72
- package/dist/cli/autoaccept-poll.js +23 -0
- package/dist/cli/drive-write-pretool.mjs +24 -1
- package/dist/cli/foreground-hog-pretool.mjs +264 -0
- package/dist/cli/ms-365-write-pretool.mjs +31 -8
- package/dist/cli/notion-write-pretool.mjs +9 -2
- package/dist/cli/skill-validate-pretool.mjs +144 -2847
- package/dist/cli/switchroom.js +986 -3131
- package/dist/host-control/main.js +216 -2863
- package/dist/vault/approvals/kernel-server.js +67 -1
- package/dist/vault/broker/server.js +98 -45
- package/package.json +1 -1
- package/profiles/coding/CLAUDE.md.hbs +2 -0
- package/profiles/default/CLAUDE.md.hbs +2 -0
- package/skills/switchroom-architecture/telegram.md +0 -1
- package/telegram-plugin/auth-snapshot-format.ts +37 -5
- package/telegram-plugin/auto-fallback-fleet.ts +29 -1
- package/telegram-plugin/bridge/bridge.ts +2 -0
- package/telegram-plugin/dist/bridge/bridge.js +51 -3
- package/telegram-plugin/dist/gateway/gateway.js +1251 -2368
- package/telegram-plugin/dist/server.js +67 -3
- package/telegram-plugin/format.ts +19 -0
- package/telegram-plugin/gateway/approval-hold.ts +21 -2
- package/telegram-plugin/gateway/auth-broker-client.ts +1 -0
- package/telegram-plugin/gateway/auth-command.ts +14 -0
- package/telegram-plugin/gateway/callback-query-handlers.ts +12 -0
- package/telegram-plugin/gateway/forward-origin.ts +235 -0
- package/telegram-plugin/gateway/gateway.ts +445 -83
- package/telegram-plugin/gateway/throttle-tier-wiring.ts +268 -0
- package/telegram-plugin/history.ts +106 -6
- package/telegram-plugin/inline-keyboard-callbacks.ts +94 -0
- package/telegram-plugin/model-unavailable.ts +61 -13
- package/telegram-plugin/outbound-field-redact.ts +69 -0
- package/telegram-plugin/render/render.ts +32 -14
- package/telegram-plugin/render/rich-render.ts +40 -32
- package/telegram-plugin/scoped-approval.ts +11 -2
- package/telegram-plugin/secret-detect/chunker.ts +18 -4
- package/telegram-plugin/secret-detect/index.ts +12 -56
- package/telegram-plugin/send-gate-degraded.test.ts +131 -0
- package/telegram-plugin/send-gate.test.ts +25 -6
- package/telegram-plugin/send-gate.ts +82 -8
- package/telegram-plugin/session-tail.ts +82 -7
- package/telegram-plugin/stream-controller.ts +3 -2
- package/telegram-plugin/subagent-watcher.ts +71 -16
- package/telegram-plugin/tests/approval-hold-outcome.test.ts +36 -5
- package/telegram-plugin/tests/auto-fallback-fleet.test.ts +72 -0
- package/telegram-plugin/tests/callback-query-handlers.test.ts +65 -0
- package/telegram-plugin/tests/forward-origin.test.ts +309 -0
- package/telegram-plugin/tests/gateway-outbound-redact.test.ts +57 -0
- package/telegram-plugin/tests/history.test.ts +272 -0
- package/telegram-plugin/tests/inbound-message-types.test.ts +5 -1
- package/telegram-plugin/tests/inline-keyboard-callbacks.test.ts +164 -0
- package/telegram-plugin/tests/operator-events-session-tail.test.ts +74 -0
- package/telegram-plugin/tests/outbound-field-redact.test.ts +107 -0
- package/telegram-plugin/tests/reaction-gate-routing.test.ts +173 -0
- package/telegram-plugin/tests/render/render-outbound-chunks.test.ts +6 -4
- package/telegram-plugin/tests/render/render.test.ts +88 -0
- package/telegram-plugin/tests/render/rich-render.test.ts +41 -22
- package/telegram-plugin/tests/scoped-approval.test.ts +27 -0
- package/telegram-plugin/tests/secret-detect-chunk-overlap.test.ts +65 -0
- package/telegram-plugin/tests/secret-detect-oauth-code.test.ts +5 -4
- package/telegram-plugin/tests/session-tail-sidecar-reap.test.ts +268 -0
- package/telegram-plugin/tests/single-mode-stream-reply.test.ts +5 -3
- package/telegram-plugin/tests/status-accent.test.ts +5 -3
- package/telegram-plugin/tests/stream-controller-chunk-cap.test.ts +20 -20
- package/telegram-plugin/tests/stream-reply-handler.test.ts +5 -2
- package/telegram-plugin/tests/subagent-watcher-fd-leak.test.ts +275 -0
- package/telegram-plugin/tests/throttle-tier-wiring.test.ts +290 -0
- package/telegram-plugin/tests/throttle-tier.test.ts +278 -0
- package/telegram-plugin/tests/worktree-watch-cwds.test.ts +215 -1
- package/telegram-plugin/throttle-tier.ts +226 -0
- package/telegram-plugin/uat/scenarios/jtbd-rich-formatting-render-dm.test.ts +8 -7
- package/telegram-plugin/worktree-watch-cwds.ts +194 -5
- package/telegram-plugin/secret-detect/secretlint-source.ts +0 -95
- package/telegram-plugin/tests/secret-detect-secretlint.test.ts +0 -105
|
@@ -0,0 +1,226 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* throttle-tier.ts — the 429 throttle tier (pure decision + notice rendering).
|
|
3
|
+
*
|
|
4
|
+
* Operator-approved behavior spec: "retry in place under 5 min, else mark +
|
|
5
|
+
* failover, honest reset messaging."
|
|
6
|
+
*
|
|
7
|
+
* A terminal TRANSIENT 429 (a `rate_limit_error` whose wording explicitly
|
|
8
|
+
* negates the account-quota reading — see `transientUpstreamSignals` in
|
|
9
|
+
* model-unavailable.ts) used to take the fully calm path: no broker state, no
|
|
10
|
+
* account attribution, a generic "🚦 Rate limited" card. That is right for a
|
|
11
|
+
* few-second burst but wrong for the minutes-long per-account throttles
|
|
12
|
+
* Anthropic emits with a "resets 8:50am (TZ)" hint: the fleet has no memory
|
|
13
|
+
* that the account is throttled (cron fires walk straight into the same 429)
|
|
14
|
+
* and the operator can't tell a throttle from a wall.
|
|
15
|
+
*
|
|
16
|
+
* This module owns the DECISION and the NOTICE TEXT; the gateway wires the
|
|
17
|
+
* side effects (broker `mark-throttled`, the Telegram send, the delayed
|
|
18
|
+
* retry nudge):
|
|
19
|
+
*
|
|
20
|
+
* - reset parseable and ≤ threshold (default 5 min) → `throttle`: stay on
|
|
21
|
+
* the account, record `throttled_until` broker-side, notify once.
|
|
22
|
+
* - reset parseable and > threshold → `failover`: escalate to the existing
|
|
23
|
+
* mark-exhausted + fleet-failover machinery with the parsed reset as the
|
|
24
|
+
* mark expiry.
|
|
25
|
+
* - reset unparseable → `throttle` with a conservative now+60s wait.
|
|
26
|
+
*
|
|
27
|
+
* Wall wording ("You've hit your limit", out_of_credits) never reaches this
|
|
28
|
+
* module — session-tail classifies those `quota-exhausted` and the gateway's
|
|
29
|
+
* existing failover path handles them unchanged.
|
|
30
|
+
*
|
|
31
|
+
* PURE — no IPC, no bot, no clock except the injected `now`.
|
|
32
|
+
*/
|
|
33
|
+
|
|
34
|
+
import { escapeMarkdown } from './card-format.js'
|
|
35
|
+
import { formatResetRelative } from './quota-check.js'
|
|
36
|
+
import { parseResetTime } from './model-unavailable.js'
|
|
37
|
+
|
|
38
|
+
// ─── Account-scoped throttle wording ─────────────────────────────────────────
|
|
39
|
+
|
|
40
|
+
/**
|
|
41
|
+
* ACCOUNT-AFFIRMING subset of `transientUpstreamSignals` (model-unavailable.ts).
|
|
42
|
+
* The throttle tier takes ACCOUNT-level actions — broker `mark-throttled`,
|
|
43
|
+
* per-account notice, retry nudge — so it must key on wording that affirms the
|
|
44
|
+
* account's OWN rate limit ("would exceed your account's rate limit",
|
|
45
|
+
* "not your account…"), NOT the server-side phrasings ("Server is temporarily
|
|
46
|
+
* limiting requests (not your usage limit)", 529 overload wording). A
|
|
47
|
+
* server-wide condition recorded as an account throttle would bench the wrong
|
|
48
|
+
* thing and restart-nudge an agent straight back into the same server-side
|
|
49
|
+
* wall. Server-side transients keep the existing calm rate-limited path
|
|
50
|
+
* (Claude Code's internal retry) untouched.
|
|
51
|
+
*/
|
|
52
|
+
export const accountScopedThrottleSignals = [
|
|
53
|
+
"would exceed your account's rate limit",
|
|
54
|
+
'would exceed your account’s rate limit',
|
|
55
|
+
// Covers both "not your account" and "not your account's" (substring).
|
|
56
|
+
'not your account',
|
|
57
|
+
]
|
|
58
|
+
|
|
59
|
+
/** True when `text` carries an EXPLICIT account-affirming throttle marker.
|
|
60
|
+
* Never throws on weird input. */
|
|
61
|
+
export function isAccountScopedThrottle(text: string): boolean {
|
|
62
|
+
if (typeof text !== 'string' || text.length === 0) return false
|
|
63
|
+
const sample = text.length > 16_384 ? text.slice(0, 16_384) : text
|
|
64
|
+
const lower = sample.toLowerCase()
|
|
65
|
+
return accountScopedThrottleSignals.some((s) => lower.includes(s))
|
|
66
|
+
}
|
|
67
|
+
|
|
68
|
+
/**
|
|
69
|
+
* Default retry-in-place ceiling: a transient 429 whose reset is within this
|
|
70
|
+
* window waits on the account instead of failing the fleet over. Override
|
|
71
|
+
* with SWITCHROOM_THROTTLE_RETRY_IN_PLACE_MAX_MS.
|
|
72
|
+
*/
|
|
73
|
+
export const THROTTLE_RETRY_IN_PLACE_MAX_MS_DEFAULT = 5 * 60_000
|
|
74
|
+
|
|
75
|
+
/**
|
|
76
|
+
* Wait applied when a transient 429 carries NO parseable reset. Anthropic's
|
|
77
|
+
* burst throttles clear in seconds; 60s is comfortably past the common case
|
|
78
|
+
* without benching the account for long on a misread.
|
|
79
|
+
*/
|
|
80
|
+
export const THROTTLE_DEFAULT_WAIT_MS = 60_000
|
|
81
|
+
|
|
82
|
+
/** Resolve the retry-in-place threshold from env (ms), else the default. */
|
|
83
|
+
export function throttleRetryInPlaceMaxMs(
|
|
84
|
+
env: NodeJS.ProcessEnv = process.env,
|
|
85
|
+
): number {
|
|
86
|
+
const raw = env.SWITCHROOM_THROTTLE_RETRY_IN_PLACE_MAX_MS
|
|
87
|
+
if (raw == null || raw === '') return THROTTLE_RETRY_IN_PLACE_MAX_MS_DEFAULT
|
|
88
|
+
const n = Number(raw)
|
|
89
|
+
return Number.isFinite(n) && n > 0 ? n : THROTTLE_RETRY_IN_PLACE_MAX_MS_DEFAULT
|
|
90
|
+
}
|
|
91
|
+
|
|
92
|
+
export type ThrottleTierDecision =
|
|
93
|
+
/** Not a transient per-account 429 — caller falls through to its
|
|
94
|
+
* existing handling (calm card / quota path). */
|
|
95
|
+
| { action: 'none' }
|
|
96
|
+
/** Stay on the account; record throttled_until broker-side; notify once. */
|
|
97
|
+
| { action: 'throttle'; throttledUntilMs: number; resetParsed: boolean }
|
|
98
|
+
/** Reset is beyond the retry-in-place threshold — escalate to the
|
|
99
|
+
* existing mark-exhausted + fleet-failover machinery. */
|
|
100
|
+
| { action: 'failover'; resetAtMs: number }
|
|
101
|
+
|
|
102
|
+
/**
|
|
103
|
+
* THE throttle-tier decision for a terminal `rate-limited` operator event.
|
|
104
|
+
* `detail` is the user-facing error text from the transcript (the source of
|
|
105
|
+
* both the transient-negation wording and the "resets …" prose).
|
|
106
|
+
*/
|
|
107
|
+
export function decideThrottleTier(opts: {
|
|
108
|
+
detail: string
|
|
109
|
+
now: number
|
|
110
|
+
/** Retry-in-place ceiling (ms). Callers pass throttleRetryInPlaceMaxMs(). */
|
|
111
|
+
thresholdMs: number
|
|
112
|
+
}): ThrottleTierDecision {
|
|
113
|
+
const { detail, now, thresholdMs } = opts
|
|
114
|
+
// Account-affirming wording ONLY — the wider transient list includes
|
|
115
|
+
// server-side phrasings (529 / "server is temporarily limiting requests")
|
|
116
|
+
// for which an account-scoped throttle would be the wrong action.
|
|
117
|
+
if (!isAccountScopedThrottle(detail)) return { action: 'none' }
|
|
118
|
+
const resetAt = parseResetTime(detail, new Date(now))
|
|
119
|
+
const resetAtMs = resetAt?.getTime()
|
|
120
|
+
if (resetAtMs == null || !Number.isFinite(resetAtMs) || resetAtMs <= now) {
|
|
121
|
+
// No usable reset — wait the conservative default in place.
|
|
122
|
+
return {
|
|
123
|
+
action: 'throttle',
|
|
124
|
+
throttledUntilMs: now + THROTTLE_DEFAULT_WAIT_MS,
|
|
125
|
+
resetParsed: false,
|
|
126
|
+
}
|
|
127
|
+
}
|
|
128
|
+
if (resetAtMs - now <= thresholdMs) {
|
|
129
|
+
return { action: 'throttle', throttledUntilMs: resetAtMs, resetParsed: true }
|
|
130
|
+
}
|
|
131
|
+
return { action: 'failover', resetAtMs }
|
|
132
|
+
}
|
|
133
|
+
|
|
134
|
+
// ─── Per-account notice cooldown ─────────────────────────────────────────────
|
|
135
|
+
|
|
136
|
+
/**
|
|
137
|
+
* Cooldown for the lightweight throttle notice, PER ACCOUNT. A burst of
|
|
138
|
+
* transient 429s (session-tail forwards every terminal error line) must
|
|
139
|
+
* produce ONE notice, not a stream — same shape and rationale as the
|
|
140
|
+
* failure-notice / all-blocked cooldowns in auto-fallback-fleet.ts. Keyed by
|
|
141
|
+
* account (not agent): the throttle is account-scoped, so every agent riding
|
|
142
|
+
* the same throttled account shares one window.
|
|
143
|
+
*/
|
|
144
|
+
export const THROTTLE_NOTICE_COOLDOWN_MS = 10 * 60_000
|
|
145
|
+
|
|
146
|
+
export interface ThrottleNoticeState {
|
|
147
|
+
/** account label → unix ms of the last notice sent for it. */
|
|
148
|
+
lastSentAtMsByAccount: Record<string, number>
|
|
149
|
+
}
|
|
150
|
+
|
|
151
|
+
export function evaluateThrottleNotice(
|
|
152
|
+
prev: ThrottleNoticeState,
|
|
153
|
+
account: string,
|
|
154
|
+
now: number,
|
|
155
|
+
cooldownMs: number = THROTTLE_NOTICE_COOLDOWN_MS,
|
|
156
|
+
): { send: boolean; next: ThrottleNoticeState } {
|
|
157
|
+
const last = prev.lastSentAtMsByAccount[account] ?? 0
|
|
158
|
+
if (now - last >= cooldownMs) {
|
|
159
|
+
return {
|
|
160
|
+
send: true,
|
|
161
|
+
next: {
|
|
162
|
+
lastSentAtMsByAccount: {
|
|
163
|
+
...prev.lastSentAtMsByAccount,
|
|
164
|
+
[account]: now,
|
|
165
|
+
},
|
|
166
|
+
},
|
|
167
|
+
}
|
|
168
|
+
}
|
|
169
|
+
return { send: false, next: prev }
|
|
170
|
+
}
|
|
171
|
+
|
|
172
|
+
// ─── Notice rendering ────────────────────────────────────────────────────────
|
|
173
|
+
|
|
174
|
+
/**
|
|
175
|
+
* The ONE lightweight operator notice for the throttle path. Deliberately not
|
|
176
|
+
* the ⚠️ model-unavailable card: nothing is exhausted and nothing failed
|
|
177
|
+
* over — the point is to say so, name the account, and name the reset.
|
|
178
|
+
*/
|
|
179
|
+
export function renderThrottleNotice(opts: {
|
|
180
|
+
/** Account label from the broker's mark-throttled response; null when the
|
|
181
|
+
* broker was unreachable and the account could not be attributed. */
|
|
182
|
+
account: string | null
|
|
183
|
+
agent: string
|
|
184
|
+
throttledUntilMs: number
|
|
185
|
+
/** True when the reset came from parsed "resets …" prose (vs the 60s default). */
|
|
186
|
+
resetParsed: boolean
|
|
187
|
+
now?: Date
|
|
188
|
+
}): string {
|
|
189
|
+
const now = opts.now ?? new Date()
|
|
190
|
+
// "resets in 3m" — same countdown dialect /usage speaks.
|
|
191
|
+
const resetStr = formatResetRelative(new Date(opts.throttledUntilMs), now)
|
|
192
|
+
const acct = opts.account ? `\`${escapeMarkdown(opts.account)}\`` : 'the active account'
|
|
193
|
+
const lines = [
|
|
194
|
+
`🚦 **Rate-limited, staying put** — ${acct} hit a transient rate limit on **${escapeMarkdown(opts.agent)}**.`,
|
|
195
|
+
`This is a short throttle, not a quota wall — ${
|
|
196
|
+
opts.resetParsed ? resetStr : `no reset given, retrying in ~60s`
|
|
197
|
+
}.`,
|
|
198
|
+
`_Staying on ${acct}; no failover needed. The turn retries automatically after the reset._`,
|
|
199
|
+
]
|
|
200
|
+
return lines.join('\n')
|
|
201
|
+
}
|
|
202
|
+
|
|
203
|
+
/**
|
|
204
|
+
* Notice for the broker's ESCALATION outcome: repeated transient 429s on one
|
|
205
|
+
* account were corroborated by a live probe as a genuine wall, so the broker
|
|
206
|
+
* ran the standard mark-exhausted + roll. The gateway that raised the
|
|
207
|
+
* mark-throttled announces it (the reactive-path doctrine — same reason the
|
|
208
|
+
* plain mark-exhausted path announces gateway-side rather than via
|
|
209
|
+
* `last_fleet_roll`), which also covers PINNED (non-fleet-active) accounts.
|
|
210
|
+
*/
|
|
211
|
+
export function renderThrottleEscalationNotice(opts: {
|
|
212
|
+
account: string | null
|
|
213
|
+
agent: string
|
|
214
|
+
/** The account the fleet/agents rolled to; null = every fallback blocked. */
|
|
215
|
+
rolledTo: string | null
|
|
216
|
+
}): string {
|
|
217
|
+
const acct = opts.account ? `\`${escapeMarkdown(opts.account)}\`` : 'the active account'
|
|
218
|
+
const head =
|
|
219
|
+
`⛔️ **Rate limit was actually a wall** — repeated 429s on ${acct} ` +
|
|
220
|
+
`(trigger: **${escapeMarkdown(opts.agent)}**) were corroborated by a live quota probe.`
|
|
221
|
+
const tail = opts.rolledTo
|
|
222
|
+
? `Marked exhausted and rolled to \`${escapeMarkdown(opts.rolledTo)}\`.`
|
|
223
|
+
: `Marked exhausted — no fallback account had quota (all blocked). ` +
|
|
224
|
+
`Use \`/auth add <label>\` to attach another subscription.`
|
|
225
|
+
return `${head}\n${tail}`
|
|
226
|
+
}
|
|
@@ -61,13 +61,14 @@ import { richRenderEnabled } from "../../render/rich-render.js";
|
|
|
61
61
|
const AGENT = "test-harness";
|
|
62
62
|
|
|
63
63
|
// The rich-render wiring (parse -> IR -> renderSafe -> sendRichMessage) is
|
|
64
|
-
//
|
|
65
|
-
// collapsible round-trip proof below only runs when
|
|
66
|
-
//
|
|
67
|
-
//
|
|
68
|
-
//
|
|
69
|
-
//
|
|
70
|
-
//
|
|
64
|
+
// ON BY DEFAULT, with `SWITCHROOM_RICH_RENDER=0` as the escape hatch. The
|
|
65
|
+
// expandable / collapsible round-trip proof below only runs when the driver
|
|
66
|
+
// creds are present AND the renderer hasn't been killed off — the same
|
|
67
|
+
// setting must hold in the gateway process under test for the renderer to
|
|
68
|
+
// actually shape the outbound message, so gating the scenario on the same
|
|
69
|
+
// check keeps it honest (no false green when the wiring isn't live). When
|
|
70
|
+
// the kill-switch is set, this scenario self-skips green exactly like the
|
|
71
|
+
// credential-less case.
|
|
71
72
|
const RICH_RENDER_ON = richRenderEnabled();
|
|
72
73
|
|
|
73
74
|
// The driver session is the load-bearing credential. Absent it, spinUp()
|
|
@@ -90,13 +90,30 @@ function defaultDeriveName(agentDir: string): string {
|
|
|
90
90
|
return leaf;
|
|
91
91
|
}
|
|
92
92
|
|
|
93
|
+
/**
|
|
94
|
+
* Two-tier owner-identity resolution shared by `ownedWorktreeCwds` and
|
|
95
|
+
* `refreshOwnedWorktreeHeartbeats`:
|
|
96
|
+
* Tier 1 — env fast path (`SWITCHROOM_AGENT_NAME`).
|
|
97
|
+
* Tier 2 — durable fallback derived from the agent's OWN directory basename.
|
|
98
|
+
* Returns "" when neither source yields a usable identity — callers MUST treat
|
|
99
|
+
* "" as fail-closed (never guess ownership).
|
|
100
|
+
*/
|
|
101
|
+
export function resolveOwnerIdentity(
|
|
102
|
+
self: string | undefined,
|
|
103
|
+
agentDir: string | null | undefined,
|
|
104
|
+
deriveName?: (agentDir: string) => string,
|
|
105
|
+
): string {
|
|
106
|
+
let resolved: string = self != null ? self : "";
|
|
107
|
+
if (resolved === "" && agentDir != null && agentDir !== "") {
|
|
108
|
+
const derive = deriveName ?? defaultDeriveName;
|
|
109
|
+
resolved = derive(agentDir) || "";
|
|
110
|
+
}
|
|
111
|
+
return resolved;
|
|
112
|
+
}
|
|
113
|
+
|
|
93
114
|
export function ownedWorktreeCwds(opts: OwnedWorktreeCwdsOptions): string[] {
|
|
94
115
|
// Tier 1: env fast path. Tier 2: durable agentDir-derived fallback.
|
|
95
|
-
|
|
96
|
-
if (resolved === "" && opts.agentDir != null && opts.agentDir !== "") {
|
|
97
|
-
const derive = opts.deriveName ?? defaultDeriveName;
|
|
98
|
-
resolved = derive(opts.agentDir) || "";
|
|
99
|
-
}
|
|
116
|
+
const resolved = resolveOwnerIdentity(opts.self, opts.agentDir, opts.deriveName);
|
|
100
117
|
|
|
101
118
|
if (resolved === "") {
|
|
102
119
|
// Both env and durable config unavailable. Keep the historical
|
|
@@ -133,3 +150,175 @@ export function ownedWorktreeCwds(opts: OwnedWorktreeCwdsOptions): string[] {
|
|
|
133
150
|
return [];
|
|
134
151
|
}
|
|
135
152
|
}
|
|
153
|
+
|
|
154
|
+
/**
|
|
155
|
+
* Default minimum interval (ms) between heartbeat writes for the same
|
|
156
|
+
* worktree record. The gateway invokes the refresh on every ~1s rescan tick;
|
|
157
|
+
* writing every tick would be needless churn. Refreshing at most every 2 min
|
|
158
|
+
* keeps every live claim's heartbeat FAR fresher than the reaper's 10-min
|
|
159
|
+
* `STALE_THRESHOLD_MS`, so a live claim never reads as stale, while a dead
|
|
160
|
+
* gateway (no ticks) lets the heartbeat age out and become reap-eligible.
|
|
161
|
+
*/
|
|
162
|
+
export const DEFAULT_HEARTBEAT_REFRESH_INTERVAL_MS = 2 * 60_000;
|
|
163
|
+
|
|
164
|
+
/** Minimal record shape needed to refresh a worktree heartbeat. */
|
|
165
|
+
export interface WorktreeHeartbeatRecord {
|
|
166
|
+
id: string;
|
|
167
|
+
ownerAgent?: string;
|
|
168
|
+
/** ISO 8601 timestamp of the record's current heartbeat (for throttling). */
|
|
169
|
+
heartbeatAt?: string;
|
|
170
|
+
}
|
|
171
|
+
|
|
172
|
+
export interface RefreshOwnedHeartbeatsOptions {
|
|
173
|
+
/** The agent's identity — `process.env.SWITCHROOM_AGENT_NAME` (fast path). */
|
|
174
|
+
self: string | undefined;
|
|
175
|
+
/** Durable, non-env identity fallback: the agent's OWN directory. */
|
|
176
|
+
agentDir?: string | null;
|
|
177
|
+
/** Host-global registry read (`listRecords` from src/worktree/registry). */
|
|
178
|
+
listRecords: () => WorktreeHeartbeatRecord[];
|
|
179
|
+
/** Advance one record's heartbeat (`touchHeartbeat` from the registry). */
|
|
180
|
+
touchHeartbeat: (id: string) => void;
|
|
181
|
+
/**
|
|
182
|
+
* Skip a touch while the record's heartbeat is younger than this (ms).
|
|
183
|
+
* Defaults to `DEFAULT_HEARTBEAT_REFRESH_INTERVAL_MS`. Set to 0 to always
|
|
184
|
+
* touch (used by tests).
|
|
185
|
+
*/
|
|
186
|
+
minRefreshIntervalMs?: number;
|
|
187
|
+
/** `Date.now` override for tests. */
|
|
188
|
+
now?: () => number;
|
|
189
|
+
/** Injectable name derivation (defaults to `path.basename`). */
|
|
190
|
+
deriveName?: (agentDir: string) => string;
|
|
191
|
+
/** Best-effort error sink. */
|
|
192
|
+
log?: (msg: string) => void;
|
|
193
|
+
}
|
|
194
|
+
|
|
195
|
+
/**
|
|
196
|
+
* Refresh the heartbeat of every worktree THIS agent owns.
|
|
197
|
+
*
|
|
198
|
+
* This is the production driver that keeps `touchHeartbeat` (previously dead —
|
|
199
|
+
* F1/H3) alive: the gateway calls it on the same rescan tick that re-derives
|
|
200
|
+
* the watched worktree cwds, so a claim held by a LIVE agent has its heartbeat
|
|
201
|
+
* advanced continuously and never trips the reaper's staleness gate. When the
|
|
202
|
+
* owning gateway dies, the ticks stop, the heartbeat ages past
|
|
203
|
+
* `STALE_THRESHOLD_MS`, and the (now fail-safe) reaper can reclaim the truly
|
|
204
|
+
* abandoned claim.
|
|
205
|
+
*
|
|
206
|
+
* Fail-CLOSED, mirroring `ownedWorktreeCwds`:
|
|
207
|
+
* - Unresolved identity ⇒ touch nothing (never guess ownership).
|
|
208
|
+
* - Ownerless records (`ownerAgent` undefined) are NEVER matched — we must
|
|
209
|
+
* not advance another agent's / an unattributable claim's heartbeat.
|
|
210
|
+
* - A registry read failure ⇒ touch nothing.
|
|
211
|
+
* - A per-record touch failure is swallowed (logged) — one bad record must
|
|
212
|
+
* not abort the rest, and this runs on the hot watch loop.
|
|
213
|
+
*
|
|
214
|
+
* @returns the number of heartbeats actually advanced this call.
|
|
215
|
+
*/
|
|
216
|
+
export function refreshOwnedWorktreeHeartbeats(
|
|
217
|
+
opts: RefreshOwnedHeartbeatsOptions,
|
|
218
|
+
): number {
|
|
219
|
+
const identity = resolveOwnerIdentity(opts.self, opts.agentDir, opts.deriveName);
|
|
220
|
+
if (identity === "") return 0; // fail-closed: never guess ownership
|
|
221
|
+
|
|
222
|
+
const nowMs = (opts.now ?? Date.now)();
|
|
223
|
+
const minInterval =
|
|
224
|
+
opts.minRefreshIntervalMs ?? DEFAULT_HEARTBEAT_REFRESH_INTERVAL_MS;
|
|
225
|
+
|
|
226
|
+
let records: WorktreeHeartbeatRecord[];
|
|
227
|
+
try {
|
|
228
|
+
records = opts.listRecords();
|
|
229
|
+
} catch {
|
|
230
|
+
return 0;
|
|
231
|
+
}
|
|
232
|
+
|
|
233
|
+
let touched = 0;
|
|
234
|
+
for (const r of records) {
|
|
235
|
+
if (r.ownerAgent !== identity) continue; // ownerless never matched
|
|
236
|
+
// Throttle: skip if the heartbeat is still comfortably fresh.
|
|
237
|
+
if (minInterval > 0 && r.heartbeatAt != null) {
|
|
238
|
+
const age = nowMs - new Date(r.heartbeatAt).getTime();
|
|
239
|
+
if (Number.isFinite(age) && age >= 0 && age < minInterval) continue;
|
|
240
|
+
}
|
|
241
|
+
try {
|
|
242
|
+
opts.touchHeartbeat(r.id);
|
|
243
|
+
touched++;
|
|
244
|
+
} catch (err) {
|
|
245
|
+
opts.log?.(
|
|
246
|
+
`worktree heartbeat refresh failed for ${r.id}: ${(err as Error).message}`,
|
|
247
|
+
);
|
|
248
|
+
}
|
|
249
|
+
}
|
|
250
|
+
return touched;
|
|
251
|
+
}
|
|
252
|
+
|
|
253
|
+
/** A registry record carrying everything both watch operations need. */
|
|
254
|
+
export type WorktreeWatchRecord = WorktreeOwnershipRecord & WorktreeHeartbeatRecord;
|
|
255
|
+
|
|
256
|
+
export interface WorktreeWatchProviderOptions {
|
|
257
|
+
/** The agent's identity — `process.env.SWITCHROOM_AGENT_NAME` (fast path). */
|
|
258
|
+
self: string | undefined;
|
|
259
|
+
/** Durable, non-env identity fallback: the agent's OWN directory. */
|
|
260
|
+
agentDir?: string | null;
|
|
261
|
+
/** Host-global registry read (`listRecords` from src/worktree/registry). */
|
|
262
|
+
listRecords: () => WorktreeWatchRecord[];
|
|
263
|
+
/** Advance one record's heartbeat (`touchHeartbeat` from the registry). */
|
|
264
|
+
touchHeartbeat: (id: string) => void;
|
|
265
|
+
/** Injectable realpath for the cwd derivation (defaults to fs.realpathSync). */
|
|
266
|
+
realpath?: (p: string) => string;
|
|
267
|
+
/** Injectable name derivation from `agentDir` (defaults to path.basename). */
|
|
268
|
+
deriveName?: (agentDir: string) => string;
|
|
269
|
+
/** Heartbeat throttle window (ms); see refreshOwnedWorktreeHeartbeats. */
|
|
270
|
+
minRefreshIntervalMs?: number;
|
|
271
|
+
/** `Date.now` override for tests. */
|
|
272
|
+
now?: () => number;
|
|
273
|
+
/** Best-effort log sink shared by both operations. */
|
|
274
|
+
log?: (msg: string) => void;
|
|
275
|
+
}
|
|
276
|
+
|
|
277
|
+
/**
|
|
278
|
+
* Build the `extraWatchCwdsProvider` closure the gateway installs on the
|
|
279
|
+
* subagent watcher.
|
|
280
|
+
*
|
|
281
|
+
* The provider is invoked on every ~1s rescan tick and does TWO things on that
|
|
282
|
+
* single tick:
|
|
283
|
+
* 1. advances the heartbeat of every worktree THIS agent owns
|
|
284
|
+
* (`refreshOwnedWorktreeHeartbeats` — the production driver that keeps
|
|
285
|
+
* `touchHeartbeat` alive; without it every claim reads "stale" 10 min
|
|
286
|
+
* after creation and the reaper's staleness guarantee collapses), AND
|
|
287
|
+
* 2. returns the set of owned worktree cwds for the watcher to also watch
|
|
288
|
+
* (`ownedWorktreeCwds`).
|
|
289
|
+
*
|
|
290
|
+
* It is extracted from the gateway (rather than inlined) SO THAT the wiring —
|
|
291
|
+
* specifically that the provider actually DRIVES heartbeats, not merely returns
|
|
292
|
+
* cwds — is under direct unit test. Deleting the heartbeat refresh here is
|
|
293
|
+
* caught by the provider's behaviour test (it would return cwds but stop
|
|
294
|
+
* advancing heartbeats), which the pre-extraction inline closure could not
|
|
295
|
+
* assert against.
|
|
296
|
+
*
|
|
297
|
+
* Both operations use the SAME two-tier identity (env fast path + agentDir
|
|
298
|
+
* fallback), so a single `agentDir` kill-switch on the caller governs both.
|
|
299
|
+
* Fully best-effort: neither operation throws out of the returned closure.
|
|
300
|
+
*/
|
|
301
|
+
export function makeWorktreeWatchProvider(
|
|
302
|
+
opts: WorktreeWatchProviderOptions,
|
|
303
|
+
): () => string[] {
|
|
304
|
+
return () => {
|
|
305
|
+
refreshOwnedWorktreeHeartbeats({
|
|
306
|
+
self: opts.self,
|
|
307
|
+
agentDir: opts.agentDir,
|
|
308
|
+
listRecords: opts.listRecords,
|
|
309
|
+
touchHeartbeat: opts.touchHeartbeat,
|
|
310
|
+
minRefreshIntervalMs: opts.minRefreshIntervalMs,
|
|
311
|
+
now: opts.now,
|
|
312
|
+
deriveName: opts.deriveName,
|
|
313
|
+
log: opts.log,
|
|
314
|
+
});
|
|
315
|
+
return ownedWorktreeCwds({
|
|
316
|
+
self: opts.self,
|
|
317
|
+
agentDir: opts.agentDir,
|
|
318
|
+
listRecords: opts.listRecords,
|
|
319
|
+
realpath: opts.realpath,
|
|
320
|
+
deriveName: opts.deriveName,
|
|
321
|
+
log: opts.log,
|
|
322
|
+
});
|
|
323
|
+
};
|
|
324
|
+
}
|
|
@@ -1,95 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Secretlint wrapper — adapts `@secretlint/core` + the recommend preset into
|
|
3
|
-
* our `Detection` shape so it can merge with the vendored pattern engine.
|
|
4
|
-
*
|
|
5
|
-
* Secretlint is async (it loads rules, applies preset config, walks the
|
|
6
|
-
* source). This module exposes `detectViaSecretlint(text)` returning a
|
|
7
|
-
* Promise. The synchronous `detectSecrets()` path in `index.ts` stays the
|
|
8
|
-
* fast default; callers that want the full engine use `detectSecretsAsync()`
|
|
9
|
-
* which fans out both and merges.
|
|
10
|
-
*
|
|
11
|
-
* Slug derivation: Secretlint rules don't give us a clean LHS (KEY=value),
|
|
12
|
-
* so we derive from the rule id (e.g. `@secretlint/secretlint-rule-slack`
|
|
13
|
-
* becomes the slug `@secretlint-rule-slack_YYYYMMDD` via `deriveSlug`'s
|
|
14
|
-
* rule_id + date fallback path).
|
|
15
|
-
*
|
|
16
|
-
* Confidence tier: Secretlint is a curated engine with checksum-validated
|
|
17
|
-
* rules for most providers, so every hit is `high`.
|
|
18
|
-
*/
|
|
19
|
-
|
|
20
|
-
import { lintSource } from '@secretlint/core'
|
|
21
|
-
import { creator as presetRecommendCreator } from '@secretlint/secretlint-rule-preset-recommend'
|
|
22
|
-
import type { Detection } from './index.js'
|
|
23
|
-
import { deriveSlug } from './slug.js'
|
|
24
|
-
import { isSuppressed } from './suppressor.js'
|
|
25
|
-
|
|
26
|
-
/**
|
|
27
|
-
* Map a single Secretlint rule id to a short rule slug used as the
|
|
28
|
-
* `Detection.rule_id`. The full `@secretlint/secretlint-rule-foo` names are
|
|
29
|
-
* long; we strip the scope/prefix to keep rule_ids readable in logs.
|
|
30
|
-
*/
|
|
31
|
-
function normalizeRuleId(secretlintRuleId: string): string {
|
|
32
|
-
// "@secretlint/secretlint-rule-slack" → "secretlint_slack"
|
|
33
|
-
// "secretlint-rule-custom-thing" → "secretlint_custom_thing"
|
|
34
|
-
const stripped = secretlintRuleId
|
|
35
|
-
.replace(/^@secretlint\//, '')
|
|
36
|
-
.replace(/^secretlint-rule-/, '')
|
|
37
|
-
.replace(/[^A-Za-z0-9]+/g, '_')
|
|
38
|
-
.replace(/_+/g, '_')
|
|
39
|
-
.replace(/^_+|_+$/g, '')
|
|
40
|
-
return `secretlint_${stripped || 'unknown'}`
|
|
41
|
-
}
|
|
42
|
-
|
|
43
|
-
export async function detectViaSecretlint(text: string): Promise<Detection[]> {
|
|
44
|
-
if (!text || text.length === 0) return []
|
|
45
|
-
|
|
46
|
-
let result
|
|
47
|
-
try {
|
|
48
|
-
result = await lintSource({
|
|
49
|
-
source: {
|
|
50
|
-
content: text,
|
|
51
|
-
filePath: 'message.txt',
|
|
52
|
-
ext: '.txt',
|
|
53
|
-
contentType: 'text',
|
|
54
|
-
},
|
|
55
|
-
options: {
|
|
56
|
-
config: {
|
|
57
|
-
rules: [
|
|
58
|
-
{
|
|
59
|
-
id: '@secretlint/secretlint-rule-preset-recommend',
|
|
60
|
-
rule: presetRecommendCreator,
|
|
61
|
-
},
|
|
62
|
-
],
|
|
63
|
-
},
|
|
64
|
-
noPhysicFilePath: true,
|
|
65
|
-
},
|
|
66
|
-
})
|
|
67
|
-
} catch {
|
|
68
|
-
// Fail-open: Secretlint crashes must never break the detector path.
|
|
69
|
-
return []
|
|
70
|
-
}
|
|
71
|
-
|
|
72
|
-
const existing = new Set<string>()
|
|
73
|
-
const out: Detection[] = []
|
|
74
|
-
for (const msg of result.messages) {
|
|
75
|
-
const [start, end] = msg.range
|
|
76
|
-
if (typeof start !== 'number' || typeof end !== 'number' || end <= start) continue
|
|
77
|
-
const matched_text = text.slice(start, end)
|
|
78
|
-
if (!matched_text) continue
|
|
79
|
-
const rule_id = normalizeRuleId(msg.ruleId)
|
|
80
|
-
const suggested_slug = deriveSlug({ rule_id }, existing)
|
|
81
|
-
existing.add(suggested_slug)
|
|
82
|
-
out.push({
|
|
83
|
-
rule_id,
|
|
84
|
-
matched_text,
|
|
85
|
-
start,
|
|
86
|
-
end,
|
|
87
|
-
confidence: 'high',
|
|
88
|
-
suppressed: isSuppressed(text, start, end),
|
|
89
|
-
suggested_slug,
|
|
90
|
-
})
|
|
91
|
-
}
|
|
92
|
-
return out
|
|
93
|
-
}
|
|
94
|
-
|
|
95
|
-
export { normalizeRuleId as __normalizeRuleId }
|
|
@@ -1,105 +0,0 @@
|
|
|
1
|
-
import { describe, it, expect } from 'vitest'
|
|
2
|
-
import {
|
|
3
|
-
detectViaSecretlint,
|
|
4
|
-
detectSecretsAsync,
|
|
5
|
-
} from '../secret-detect/index.js'
|
|
6
|
-
|
|
7
|
-
// All three fixtures are assembled at runtime so the source file never
|
|
8
|
-
// contains a contiguous token pattern. Matches the corresponding
|
|
9
|
-
// secretlint regex when evaluated; evades GitHub Push Protection's
|
|
10
|
-
// static-text scan. See CLAUDE.md "Secrets in tests".
|
|
11
|
-
const SLACK_FIXTURE = ['xoxb', '0000000000', '0000000000000', 'FIXTURE0NOTAREALTOKEN000'].join('-')
|
|
12
|
-
const GITHUB_FIXTURE = 'ghp' + '_' + '16C7e42F292c6912E7710c838347Ae178B4a'
|
|
13
|
-
const NPM_FIXTURE = 'npm' + '_' + 'AbCdEfGhIjKlMnOpQrStUvWxYz0123456789'
|
|
14
|
-
|
|
15
|
-
describe('secretlint-source.detectViaSecretlint', () => {
|
|
16
|
-
it('catches a realistic-looking Slack bot token', async () => {
|
|
17
|
-
const text = 'Slack: ' + SLACK_FIXTURE
|
|
18
|
-
const hits = await detectViaSecretlint(text)
|
|
19
|
-
expect(hits.length).toBeGreaterThan(0)
|
|
20
|
-
const slack = hits.find((h) => h.rule_id.includes('slack'))
|
|
21
|
-
expect(slack).toBeDefined()
|
|
22
|
-
expect(slack!.confidence).toBe('high')
|
|
23
|
-
expect(slack!.suppressed).toBe(false)
|
|
24
|
-
// rule_id normalized from @secretlint/secretlint-rule-slack → secretlint_slack
|
|
25
|
-
expect(slack!.rule_id).toMatch(/^secretlint_/)
|
|
26
|
-
expect(slack!.rule_id).toContain('slack')
|
|
27
|
-
// Matched text should be the actual token bytes.
|
|
28
|
-
expect(slack!.matched_text.startsWith('xoxb-')).toBe(true)
|
|
29
|
-
// Slug derived from rule_id + date (rule_id fallback path).
|
|
30
|
-
expect(slack!.suggested_slug).toMatch(/^secretlint_slack_\d{8}/)
|
|
31
|
-
})
|
|
32
|
-
|
|
33
|
-
it('catches a GitHub personal access token', async () => {
|
|
34
|
-
const text = 'token=' + GITHUB_FIXTURE + ' rest of message'
|
|
35
|
-
const hits = await detectViaSecretlint(text)
|
|
36
|
-
const gh = hits.find((h) => h.rule_id.includes('github'))
|
|
37
|
-
expect(gh).toBeDefined()
|
|
38
|
-
expect(gh!.confidence).toBe('high')
|
|
39
|
-
expect(gh!.matched_text.startsWith('ghp_')).toBe(true)
|
|
40
|
-
expect(gh!.rule_id).toContain('github')
|
|
41
|
-
})
|
|
42
|
-
|
|
43
|
-
it('catches an NPM access token', async () => {
|
|
44
|
-
const text = 'NPM_TOKEN=' + NPM_FIXTURE
|
|
45
|
-
const hits = await detectViaSecretlint(text)
|
|
46
|
-
const npm = hits.find((h) => h.rule_id.includes('npm'))
|
|
47
|
-
expect(npm).toBeDefined()
|
|
48
|
-
expect(npm!.confidence).toBe('high')
|
|
49
|
-
expect(npm!.matched_text.startsWith('npm_')).toBe(true)
|
|
50
|
-
})
|
|
51
|
-
|
|
52
|
-
it('returns empty for empty input', async () => {
|
|
53
|
-
expect(await detectViaSecretlint('')).toEqual([])
|
|
54
|
-
})
|
|
55
|
-
|
|
56
|
-
it('returns empty for text with no secrets', async () => {
|
|
57
|
-
expect(await detectViaSecretlint('hello how are you today')).toEqual([])
|
|
58
|
-
})
|
|
59
|
-
|
|
60
|
-
it('marks nearby test/mock markers as suppressed', async () => {
|
|
61
|
-
const text = 'test example: ' + SLACK_FIXTURE
|
|
62
|
-
const hits = await detectViaSecretlint(text)
|
|
63
|
-
const slack = hits.find((h) => h.rule_id.includes('slack'))
|
|
64
|
-
expect(slack).toBeDefined()
|
|
65
|
-
expect(slack!.suppressed).toBe(true)
|
|
66
|
-
})
|
|
67
|
-
})
|
|
68
|
-
|
|
69
|
-
describe('detectSecretsAsync merge', () => {
|
|
70
|
-
it('merges Secretlint hits with vendored pattern hits, deduped by range', async () => {
|
|
71
|
-
// Slack token that matches both the vendored anchored pattern AND Secretlint.
|
|
72
|
-
const text = 'a ' + SLACK_FIXTURE + ' end'
|
|
73
|
-
const hits = await detectSecretsAsync(text)
|
|
74
|
-
// One entry for the Slack token — not two. Vendored wins on ties.
|
|
75
|
-
const slackHits = hits.filter(
|
|
76
|
-
(h) => h.matched_text.startsWith('xoxb-'),
|
|
77
|
-
)
|
|
78
|
-
expect(slackHits).toHaveLength(1)
|
|
79
|
-
// Vendored rule id wins on exact-range ties (listed first in merge).
|
|
80
|
-
expect(slackHits[0]!.rule_id).toBe('slack_token')
|
|
81
|
-
})
|
|
82
|
-
|
|
83
|
-
it('detects a Shopify token via the async (Secretlint-augmented) path', async () => {
|
|
84
|
-
// Shopify is now ALSO a vendored PROVIDER_PATTERN (shopify_shared_secret),
|
|
85
|
-
// so on this span the merge prefers the vendored high hit over the
|
|
86
|
-
// Secretlint one — both are valid Shopify classifications. Secretlint
|
|
87
|
-
// remains the fallback for the long tail of providers we don't vendor;
|
|
88
|
-
// this asserts the async path still detects + classifies the token.
|
|
89
|
-
const text = 'SHOPIFY=shpss_1234567890abcdef1234567890abcdef and go'
|
|
90
|
-
const hits = await detectSecretsAsync(text)
|
|
91
|
-
const shopify = hits.find((h) => h.matched_text.startsWith('shpss_'))
|
|
92
|
-
expect(shopify).toBeDefined()
|
|
93
|
-
expect(shopify!.rule_id).toMatch(/shopify/)
|
|
94
|
-
expect(shopify!.confidence).toBe('high')
|
|
95
|
-
})
|
|
96
|
-
|
|
97
|
-
it('produces unique slugs across the merged detection list', async () => {
|
|
98
|
-
const text =
|
|
99
|
-
'tok1=' + GITHUB_FIXTURE +
|
|
100
|
-
' and tok2=' + SLACK_FIXTURE
|
|
101
|
-
const hits = await detectSecretsAsync(text)
|
|
102
|
-
const slugs = hits.map((h) => h.suggested_slug)
|
|
103
|
-
expect(new Set(slugs).size).toBe(slugs.length)
|
|
104
|
-
})
|
|
105
|
-
})
|