@aixle/insights 0.2.0 → 0.2.1
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/README.md +91 -15
- package/dist/auth/credentials.d.ts +7 -1
- package/dist/auth/credentials.js +90 -17
- package/dist/auth/flow.js +1 -0
- package/dist/cli.js +1 -1
- package/dist/collect-cursor-payloads.d.ts +1 -0
- package/dist/collect-cursor-payloads.js +8 -5
- package/dist/cursor-payload-contract.js +6 -0
- package/dist/cursor-settings.d.ts +9 -4
- package/dist/cursor-settings.js +80 -10
- package/dist/health.d.ts +1 -1
- package/dist/health.js +1 -1
- package/dist/hooks/cursor-hooks-reader.d.ts +2 -0
- package/dist/hooks/cursor-hooks-reader.js +10 -3
- package/dist/lib/client.d.ts +7 -0
- package/dist/lib/client.js +17 -0
- package/dist/lib/config.d.ts +2 -2
- package/dist/lib/config.js +32 -19
- package/dist/lib/parse-error.d.ts +21 -0
- package/dist/lib/parse-error.js +25 -0
- package/dist/lib/project-resolver.d.ts +5 -4
- package/dist/lib/project-resolver.js +57 -11
- package/dist/lib/repo-path-safety.d.ts +35 -0
- package/dist/lib/repo-path-safety.js +102 -0
- package/dist/lib/spawn-arg-safety.d.ts +25 -0
- package/dist/lib/spawn-arg-safety.js +49 -0
- package/dist/readers/claude.d.ts +59 -6
- package/dist/readers/claude.js +184 -3
- package/dist/readers/cursor.d.ts +6 -3
- package/dist/readers/cursor.js +125 -18
- package/dist/risk-scanner.js +7 -0
- package/dist/server.d.ts +3 -3
- package/dist/server.js +77 -66
- package/dist/state.js +42 -34
- package/dist/sync.d.ts +11 -0
- package/dist/sync.js +93 -60
- package/package.json +7 -3
package/dist/health.d.ts
CHANGED
|
@@ -26,6 +26,6 @@ export interface HealthSnapshot {
|
|
|
26
26
|
/** Pick the freshest persisted operator block by `last_sync_at` lexicographic (ISO-safe). */
|
|
27
27
|
export declare function mergePersistedOperators(snapshots: McpOperatorState[]): McpOperatorState | null;
|
|
28
28
|
export declare function buildHealthSnapshot(): Promise<HealthSnapshot>;
|
|
29
|
-
/** MCP `
|
|
29
|
+
/** MCP `aixle_insights_status` JSON — single source shared with CLI health. */
|
|
30
30
|
export declare function healthSnapshotToStatusPayload(snapshot: HealthSnapshot): Record<string, unknown>;
|
|
31
31
|
export declare function formatHealthForCli(snapshot: HealthSnapshot): string;
|
package/dist/health.js
CHANGED
|
@@ -133,7 +133,7 @@ export async function buildHealthSnapshot() {
|
|
|
133
133
|
};
|
|
134
134
|
}
|
|
135
135
|
}
|
|
136
|
-
/** MCP `
|
|
136
|
+
/** MCP `aixle_insights_status` JSON — single source shared with CLI health. */
|
|
137
137
|
export function healthSnapshotToStatusPayload(snapshot) {
|
|
138
138
|
const proc = snapshot.process;
|
|
139
139
|
const pers = snapshot.persisted;
|
|
@@ -8,6 +8,8 @@ export interface ProcessHooksQueueParams {
|
|
|
8
8
|
state: State;
|
|
9
9
|
host: string;
|
|
10
10
|
token: string;
|
|
11
|
+
/** Mirrors StoredCredentials.insecureHttpAllowed — set when `init --insecure` was used for this host. */
|
|
12
|
+
allowInsecureHttp?: boolean;
|
|
11
13
|
/** Called when a 429 is received. */
|
|
12
14
|
on429: (retryAfter: number, quotaExceeded: boolean) => void;
|
|
13
15
|
/** If true, skip events already in state.sessions. */
|
|
@@ -6,9 +6,16 @@ import { markSessionSent } from "../state.js";
|
|
|
6
6
|
import { postEvent } from "../client.js";
|
|
7
7
|
import { mcpLog } from "../log.js";
|
|
8
8
|
import { shouldIngestHookEvent, mapHookEventToPayload, hookDedupeKey, warnOnCursorVersion, CURSOR_HOOK_STATE_PREFIX, } from "./cursor-hooks-mapper.js";
|
|
9
|
+
import { isRepoPathWithinRoot, normalizeRepoPathCandidate } from "../lib/repo-path-safety.js";
|
|
9
10
|
export { CURSOR_HOOK_STATE_PREFIX };
|
|
11
|
+
/**
|
|
12
|
+
* `workspace` is `workspace_roots[0]` from the on-disk hooks queue — an
|
|
13
|
+
* arbitrary JSON string. A plain prefix match would accept
|
|
14
|
+
* `<scopeDir>/../../elsewhere` (DB90DV-547).
|
|
15
|
+
*/
|
|
10
16
|
function isUnderScopeDir(workspace, scopeDir) {
|
|
11
|
-
|
|
17
|
+
const normalized = normalizeRepoPathCandidate(workspace);
|
|
18
|
+
return normalized !== null && isRepoPathWithinRoot(normalized, scopeDir);
|
|
12
19
|
}
|
|
13
20
|
/**
|
|
14
21
|
* The forwarder redacts the home directory to "~" in workspace_roots for
|
|
@@ -47,7 +54,7 @@ function rewriteQueueKeepingLines(queuePath, allLines, keepIndices) {
|
|
|
47
54
|
* Partial failure: lines that succeeded are removed; lines that failed stay.
|
|
48
55
|
*/
|
|
49
56
|
export async function processHooksQueue(params) {
|
|
50
|
-
const { queuePath, scopeDir, host, token, on429, skipSeen = true, resolveProjectId, verbose = false, } = params;
|
|
57
|
+
const { queuePath, scopeDir, host, token, allowInsecureHttp = false, on429, skipSeen = true, resolveProjectId, verbose = false, } = params;
|
|
51
58
|
if (!existsSync(queuePath)) {
|
|
52
59
|
return { sent: 0, failed: 0, skipped: 0, state: params.state };
|
|
53
60
|
}
|
|
@@ -97,7 +104,7 @@ export async function processHooksQueue(params) {
|
|
|
97
104
|
projectId = await resolveProjectId(workspace);
|
|
98
105
|
}
|
|
99
106
|
const payload = mapHookEventToPayload(event, projectId);
|
|
100
|
-
const ok = await postEvent(payload, host, token, { on429 });
|
|
107
|
+
const ok = await postEvent(payload, host, token, { on429, allowInsecureHttp });
|
|
101
108
|
if (ok) {
|
|
102
109
|
totalSent++;
|
|
103
110
|
stateMut = markSessionSent(stateMut, dedupeKey, 0);
|
package/dist/lib/client.d.ts
CHANGED
|
@@ -8,6 +8,13 @@ export interface IngestPayload {
|
|
|
8
8
|
[key: string]: unknown;
|
|
9
9
|
}
|
|
10
10
|
export interface PostEventOptions {
|
|
11
|
+
/**
|
|
12
|
+
* Skip the HTTPS-or-loopback gate for this host. Only ever set this to
|
|
13
|
+
* `true` when the caller has independently confirmed the user explicitly
|
|
14
|
+
* consented via `init --insecure` for this exact credential (see
|
|
15
|
+
* `StoredCredentials.insecureHttpAllowed`). Defaults to `false`.
|
|
16
|
+
*/
|
|
17
|
+
allowInsecureHttp?: boolean;
|
|
11
18
|
/** Override default console.error on non-ok HTTP response. */
|
|
12
19
|
onHttpError?: (status: number, statusText: string, body: string) => void;
|
|
13
20
|
/** Override default console.error on network-level failure. */
|
package/dist/lib/client.js
CHANGED
|
@@ -1,3 +1,4 @@
|
|
|
1
|
+
import { evaluateTransportSecurity } from "./transport-security.js";
|
|
1
2
|
/**
|
|
2
3
|
* POST a single event payload to the db90 ingest endpoint.
|
|
3
4
|
*
|
|
@@ -10,6 +11,22 @@
|
|
|
10
11
|
* Never throws — callers can rely on Promise.allSettled-style aggregation.
|
|
11
12
|
*/
|
|
12
13
|
export async function postEvent(payload, host, token, options = {}) {
|
|
14
|
+
const transportSecurity = evaluateTransportSecurity(host, {
|
|
15
|
+
allowInsecureHttp: options.allowInsecureHttp === true,
|
|
16
|
+
label: "DB90 ingest host",
|
|
17
|
+
});
|
|
18
|
+
if (!transportSecurity.ok) {
|
|
19
|
+
// Deliberately does NOT call options.onNetworkError/onHttpError: the retry
|
|
20
|
+
// wrapper in src/client.ts treats those as transient and retries with
|
|
21
|
+
// backoff. A scheme rejection is permanent — retrying wastes ~21s per
|
|
22
|
+
// event for nothing. Bare console.error mirrors this file's existing
|
|
23
|
+
// unrecoverable-failure logging style.
|
|
24
|
+
console.error(`Blocked event send — ${transportSecurity.error}`);
|
|
25
|
+
return false;
|
|
26
|
+
}
|
|
27
|
+
if (transportSecurity.warning) {
|
|
28
|
+
console.error(`Warning: ${transportSecurity.warning}`);
|
|
29
|
+
}
|
|
13
30
|
const url = `${host.replace(/\/$/, "")}/api/v1/ingest/events`;
|
|
14
31
|
const headers = {
|
|
15
32
|
"Content-Type": "application/json",
|
package/dist/lib/config.d.ts
CHANGED
|
@@ -12,8 +12,8 @@ export interface BaseConfig {
|
|
|
12
12
|
* Load a connector's `config.json` from disk. Returns `{}` on missing or
|
|
13
13
|
* malformed files — callers fall back to env vars / CLI flags / defaults.
|
|
14
14
|
*
|
|
15
|
-
* @param configDir Directory containing `config.json`, typically the
|
|
16
|
-
*
|
|
15
|
+
* @param configDir Directory containing `config.json`, typically the app home directory
|
|
16
|
+
* (`~/.aixle-insights`, or `AIXLE_INSIGHTS_HOME` when set).
|
|
17
17
|
* @param parsePricing Optional callback that extracts a connector-specific
|
|
18
18
|
* pricing shape from the raw parsed JSON. Returns
|
|
19
19
|
* `undefined` when the pricing block is missing or invalid.
|
package/dist/lib/config.js
CHANGED
|
@@ -1,11 +1,13 @@
|
|
|
1
1
|
import { readFileSync } from "node:fs";
|
|
2
2
|
import { join } from "node:path";
|
|
3
|
+
import { mcpLog } from "../log.js";
|
|
4
|
+
import { describeReadFailure } from "./parse-error.js";
|
|
3
5
|
/**
|
|
4
6
|
* Load a connector's `config.json` from disk. Returns `{}` on missing or
|
|
5
7
|
* malformed files — callers fall back to env vars / CLI flags / defaults.
|
|
6
8
|
*
|
|
7
|
-
* @param configDir Directory containing `config.json`, typically the
|
|
8
|
-
*
|
|
9
|
+
* @param configDir Directory containing `config.json`, typically the app home directory
|
|
10
|
+
* (`~/.aixle-insights`, or `AIXLE_INSIGHTS_HOME` when set).
|
|
9
11
|
* @param parsePricing Optional callback that extracts a connector-specific
|
|
10
12
|
* pricing shape from the raw parsed JSON. Returns
|
|
11
13
|
* `undefined` when the pricing block is missing or invalid.
|
|
@@ -15,25 +17,36 @@ import { join } from "node:path";
|
|
|
15
17
|
*/
|
|
16
18
|
export function loadBaseConfig(configDir, parsePricing) {
|
|
17
19
|
const configPath = join(configDir, "config.json");
|
|
20
|
+
let parsed;
|
|
18
21
|
try {
|
|
19
|
-
|
|
20
|
-
|
|
21
|
-
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
};
|
|
27
|
-
if (parsePricing) {
|
|
28
|
-
const pricing = parsePricing(obj);
|
|
29
|
-
if (pricing !== undefined)
|
|
30
|
-
result.pricing = pricing;
|
|
31
|
-
}
|
|
32
|
-
return result;
|
|
22
|
+
parsed = JSON.parse(readFileSync(configPath, "utf-8"));
|
|
23
|
+
}
|
|
24
|
+
catch (err) {
|
|
25
|
+
const code = err?.code;
|
|
26
|
+
if (code !== "ENOENT") {
|
|
27
|
+
// Config file exists but is not valid JSON — distinguishes tampering from "never created".
|
|
28
|
+
// ENOENT stays silent: this file is optional and most users never create it.
|
|
29
|
+
mcpLog.warn("config_parse_failed", { path: configPath, ...describeReadFailure(err) }, false);
|
|
33
30
|
}
|
|
31
|
+
return {};
|
|
32
|
+
}
|
|
33
|
+
// Valid JSON, but not a config object. Arrays are rejected explicitly because
|
|
34
|
+
// `typeof [] === "object"` would otherwise let them reach the happy path and be handed
|
|
35
|
+
// to `parsePricing`. Previously every non-object fell through silently. (DB90DV-699)
|
|
36
|
+
if (typeof parsed !== "object" || parsed === null || Array.isArray(parsed)) {
|
|
37
|
+
mcpLog.warn("config_parse_failed", { path: configPath, reason: "invalid_shape" }, false);
|
|
38
|
+
return {};
|
|
34
39
|
}
|
|
35
|
-
|
|
36
|
-
|
|
40
|
+
const obj = parsed;
|
|
41
|
+
const result = {
|
|
42
|
+
token: typeof obj.token === "string" ? obj.token : undefined,
|
|
43
|
+
host: typeof obj.host === "string" ? obj.host : undefined,
|
|
44
|
+
project_id: typeof obj.project_id === "string" ? obj.project_id : undefined,
|
|
45
|
+
};
|
|
46
|
+
if (parsePricing) {
|
|
47
|
+
const pricing = parsePricing(obj);
|
|
48
|
+
if (pricing !== undefined)
|
|
49
|
+
result.pricing = pricing;
|
|
37
50
|
}
|
|
38
|
-
return
|
|
51
|
+
return result;
|
|
39
52
|
}
|
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
export type ReadFailureReason = "invalid_json" | "unreadable";
|
|
2
|
+
/**
|
|
3
|
+
* Classifies a caught error from `JSON.parse(readFileSync(...))` (or a parsed keychain
|
|
4
|
+
* payload) into a log-safe reason + error string.
|
|
5
|
+
*
|
|
6
|
+
* V8's `JSON.parse` throws a `SyntaxError` whose `.message` can embed a prefix (or, for a
|
|
7
|
+
* short enough input, the entirety) of the unparsed content — e.g.
|
|
8
|
+
* `JSON.parse("example_local_fixture_1234567890")` produces
|
|
9
|
+
* `Unexpected token 'e', "example_lo"... is not valid JSON`. Logging that message would
|
|
10
|
+
* leak exactly the secret content the parse-failure events exist to describe without
|
|
11
|
+
* exposing (see `credentials_parse_failed` / `credentials_keytar_parse_failed` /
|
|
12
|
+
* `config_parse_failed` / `state_parse_failed`). So for a `SyntaxError` this reports only
|
|
13
|
+
* the error name, never `.message`.
|
|
14
|
+
*
|
|
15
|
+
* Any other error (fs I/O — `EACCES`, `EISDIR`, etc.) is reported as `unreadable` using its
|
|
16
|
+
* errno `code`, which never contains file content and is more actionable than a bare name.
|
|
17
|
+
*/
|
|
18
|
+
export declare function describeReadFailure(err: unknown): {
|
|
19
|
+
reason: ReadFailureReason;
|
|
20
|
+
error: string;
|
|
21
|
+
};
|
|
@@ -0,0 +1,25 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Classifies a caught error from `JSON.parse(readFileSync(...))` (or a parsed keychain
|
|
3
|
+
* payload) into a log-safe reason + error string.
|
|
4
|
+
*
|
|
5
|
+
* V8's `JSON.parse` throws a `SyntaxError` whose `.message` can embed a prefix (or, for a
|
|
6
|
+
* short enough input, the entirety) of the unparsed content — e.g.
|
|
7
|
+
* `JSON.parse("example_local_fixture_1234567890")` produces
|
|
8
|
+
* `Unexpected token 'e', "example_lo"... is not valid JSON`. Logging that message would
|
|
9
|
+
* leak exactly the secret content the parse-failure events exist to describe without
|
|
10
|
+
* exposing (see `credentials_parse_failed` / `credentials_keytar_parse_failed` /
|
|
11
|
+
* `config_parse_failed` / `state_parse_failed`). So for a `SyntaxError` this reports only
|
|
12
|
+
* the error name, never `.message`.
|
|
13
|
+
*
|
|
14
|
+
* Any other error (fs I/O — `EACCES`, `EISDIR`, etc.) is reported as `unreadable` using its
|
|
15
|
+
* errno `code`, which never contains file content and is more actionable than a bare name.
|
|
16
|
+
*/
|
|
17
|
+
export function describeReadFailure(err) {
|
|
18
|
+
if (err instanceof SyntaxError) {
|
|
19
|
+
return { reason: "invalid_json", error: "SyntaxError" };
|
|
20
|
+
}
|
|
21
|
+
const code = err?.code;
|
|
22
|
+
if (code)
|
|
23
|
+
return { reason: "unreadable", error: code };
|
|
24
|
+
return { reason: "unreadable", error: err instanceof Error ? err.name : "unknown_error" };
|
|
25
|
+
}
|
|
@@ -6,8 +6,8 @@ export interface LookupResult {
|
|
|
6
6
|
project_id: string;
|
|
7
7
|
name: string;
|
|
8
8
|
}
|
|
9
|
-
export declare function resolveProjectId(flagValue: string | undefined, configValue: string | undefined, host: string, token: string, verbose: boolean): Promise<ProjectResolution>;
|
|
10
|
-
export declare function resolveProjectIdForRepoPath(repoPath: string, host: string, token: string, verbose: boolean): Promise<ProjectResolution>;
|
|
9
|
+
export declare function resolveProjectId(flagValue: string | undefined, configValue: string | undefined, host: string, token: string, verbose: boolean, allowInsecureHttp?: boolean): Promise<ProjectResolution>;
|
|
10
|
+
export declare function resolveProjectIdForRepoPath(repoPath: string, host: string, token: string, verbose: boolean, allowInsecureHttp?: boolean): Promise<ProjectResolution>;
|
|
11
11
|
export declare function getGitRemote(verbose: boolean): string | null;
|
|
12
12
|
export declare function getGitRemoteForPath(repoPath: string, verbose: boolean): string | null;
|
|
13
13
|
/**
|
|
@@ -24,7 +24,7 @@ export declare function canonicalizeGitRemote(remote: string, verbose: boolean):
|
|
|
24
24
|
* git remote. Expand to remotes the API lookup understands (`Project.normalize_git_remote`).
|
|
25
25
|
*/
|
|
26
26
|
export declare function repoNameToGitRemoteCandidates(repoName: string): string[];
|
|
27
|
-
export declare function lookupProjectByRepoName(repoName: string, host: string, token: string, verbose: boolean): Promise<LookupResult | "not-found" | null>;
|
|
27
|
+
export declare function lookupProjectByRepoName(repoName: string, host: string, token: string, verbose: boolean, allowInsecureHttp?: boolean): Promise<LookupResult | "not-found" | null>;
|
|
28
28
|
/** Payload shape shared by db90-cursor and telemetry-mcp commit mappers. */
|
|
29
29
|
export interface CommitAttributionPayload {
|
|
30
30
|
event_type?: string;
|
|
@@ -44,5 +44,6 @@ export declare function enrichCommitProjectAttribution(payloads: CommitAttributi
|
|
|
44
44
|
host: string;
|
|
45
45
|
token: string;
|
|
46
46
|
verbose?: boolean;
|
|
47
|
+
allowInsecureHttp?: boolean;
|
|
47
48
|
}): Promise<void>;
|
|
48
|
-
export declare function lookupProjectByRemote(gitRemote: string, host: string, token: string, verbose: boolean): Promise<LookupResult | "not-found" | null>;
|
|
49
|
+
export declare function lookupProjectByRemote(gitRemote: string, host: string, token: string, verbose: boolean, allowInsecureHttp?: boolean): Promise<LookupResult | "not-found" | null>;
|
|
@@ -1,9 +1,12 @@
|
|
|
1
1
|
import { execFileSync } from "node:child_process";
|
|
2
|
+
import { isSafeSshHost } from "./spawn-arg-safety.js";
|
|
3
|
+
import { safeGitRepoPath } from "./repo-path-safety.js";
|
|
4
|
+
import { evaluateTransportSecurity } from "./transport-security.js";
|
|
2
5
|
/** Coerce empty string to undefined so "" is treated as "not set" */
|
|
3
6
|
function coerce(val) {
|
|
4
7
|
return val === "" ? undefined : val;
|
|
5
8
|
}
|
|
6
|
-
export async function resolveProjectId(flagValue, configValue, host, token, verbose) {
|
|
9
|
+
export async function resolveProjectId(flagValue, configValue, host, token, verbose, allowInsecureHttp = false) {
|
|
7
10
|
const flag = coerce(flagValue);
|
|
8
11
|
const config = coerce(configValue);
|
|
9
12
|
if (flag !== undefined)
|
|
@@ -13,18 +16,18 @@ export async function resolveProjectId(flagValue, configValue, host, token, verb
|
|
|
13
16
|
const gitRemote = getGitRemote(verbose);
|
|
14
17
|
if (gitRemote === null)
|
|
15
18
|
return { projectId: null, source: "none" };
|
|
16
|
-
const result = await lookupProjectByRemote(gitRemote, host, token, verbose);
|
|
19
|
+
const result = await lookupProjectByRemote(gitRemote, host, token, verbose, allowInsecureHttp);
|
|
17
20
|
if (result === "not-found")
|
|
18
21
|
return { projectId: null, source: "auto-detect-not-found" };
|
|
19
22
|
if (result !== null)
|
|
20
23
|
return { projectId: result.project_id, source: "auto-detect" };
|
|
21
24
|
return { projectId: null, source: "none" };
|
|
22
25
|
}
|
|
23
|
-
export async function resolveProjectIdForRepoPath(repoPath, host, token, verbose) {
|
|
26
|
+
export async function resolveProjectIdForRepoPath(repoPath, host, token, verbose, allowInsecureHttp = false) {
|
|
24
27
|
const gitRemote = getGitRemoteForPath(repoPath, verbose);
|
|
25
28
|
if (gitRemote === null)
|
|
26
29
|
return { projectId: null, source: "none" };
|
|
27
|
-
const result = await lookupProjectByRemote(gitRemote, host, token, verbose);
|
|
30
|
+
const result = await lookupProjectByRemote(gitRemote, host, token, verbose, allowInsecureHttp);
|
|
28
31
|
if (result === "not-found")
|
|
29
32
|
return { projectId: null, source: "auto-detect-not-found" };
|
|
30
33
|
if (result !== null)
|
|
@@ -47,8 +50,19 @@ export function getGitRemote(verbose) {
|
|
|
47
50
|
}
|
|
48
51
|
}
|
|
49
52
|
export function getGitRemoteForPath(repoPath, verbose) {
|
|
53
|
+
// The spawn boundary. `repoPath` is untrusted (Cursor workspace.json, a
|
|
54
|
+
// composer uri.fsPath, a hook workspace_root, or a Claude transcript cwd), so
|
|
55
|
+
// it must resolve to a real directory before git reads its .git/config.
|
|
56
|
+
// Supersedes the isSafeSpawnPathArg check from DB90DV-546, which this
|
|
57
|
+
// subsumes. See DB90DV-547.
|
|
58
|
+
const safePath = safeGitRepoPath(repoPath);
|
|
59
|
+
if (safePath === null) {
|
|
60
|
+
if (verbose)
|
|
61
|
+
console.log(`[verbose] Refusing git for unsafe repo path: ${repoPath}`);
|
|
62
|
+
return null;
|
|
63
|
+
}
|
|
50
64
|
try {
|
|
51
|
-
const out = execFileSync("git", ["-C",
|
|
65
|
+
const out = execFileSync("git", ["-C", safePath, "remote", "get-url", "origin"], {
|
|
52
66
|
encoding: "utf-8",
|
|
53
67
|
stdio: ["ignore", "pipe", "pipe"],
|
|
54
68
|
timeout: 5000,
|
|
@@ -74,12 +88,21 @@ export function canonicalizeGitRemote(remote, verbose) {
|
|
|
74
88
|
if (!trimmed)
|
|
75
89
|
return remote;
|
|
76
90
|
const scp = trimmed.match(/^([\w.-]+)@([^:/]+):(.+)$/);
|
|
91
|
+
/* eslint-disable-next-line security/detect-unsafe-regex -- Flagged only
|
|
92
|
+
because safe-regex counts `?` as a repetition. Every group is separated by
|
|
93
|
+
a literal delimiter (`@`, `:`, `/`) that its neighbours exclude, so there
|
|
94
|
+
is no backtracking ambiguity. Input is a git remote URL, bounded length. */
|
|
77
95
|
const sshUrl = trimmed.match(/^ssh:\/\/(?:([\w.-]+)@)?([^:/]+)(?::\d+)?\/(.+)$/i);
|
|
78
96
|
const host = scp?.[2] ?? sshUrl?.[2];
|
|
79
|
-
|
|
97
|
+
// An unvalidated host would be parsed by ssh as an option (DB90DV-546); an
|
|
98
|
+
// unvalidated `resolved` would be spliced back into the remote and sent to
|
|
99
|
+
// the lookup endpoint. Both fail open — the remote is returned unchanged.
|
|
100
|
+
if (!host || !isSafeSshHost(host))
|
|
80
101
|
return trimmed;
|
|
81
102
|
const resolved = resolveSshHostName(host, verbose);
|
|
82
|
-
if (!resolved || resolved
|
|
103
|
+
if (!resolved || !isSafeSshHost(resolved))
|
|
104
|
+
return trimmed;
|
|
105
|
+
if (resolved.toLowerCase() === host.toLowerCase())
|
|
83
106
|
return trimmed;
|
|
84
107
|
if (verbose)
|
|
85
108
|
console.log(`[verbose] Resolved SSH host alias ${host} -> ${resolved}`);
|
|
@@ -89,6 +112,13 @@ export function canonicalizeGitRemote(remote, verbose) {
|
|
|
89
112
|
return `ssh://${user}${resolved}/${sshUrl[3]}`;
|
|
90
113
|
}
|
|
91
114
|
function resolveSshHostName(host, verbose) {
|
|
115
|
+
// Defense in depth: the only caller already checks, but this function is the
|
|
116
|
+
// spawn boundary and must not depend on callers getting it right.
|
|
117
|
+
if (!isSafeSshHost(host)) {
|
|
118
|
+
if (verbose)
|
|
119
|
+
console.log(`[verbose] Refusing ssh -G for option-shaped host: ${host}`);
|
|
120
|
+
return null;
|
|
121
|
+
}
|
|
92
122
|
try {
|
|
93
123
|
const out = execFileSync("ssh", ["-G", host], {
|
|
94
124
|
encoding: "utf-8",
|
|
@@ -128,12 +158,17 @@ export function repoNameToGitRemoteCandidates(repoName) {
|
|
|
128
158
|
if (trimmed.includes("://") || trimmed.includes("@")) {
|
|
129
159
|
return [trimmed];
|
|
130
160
|
}
|
|
161
|
+
/* eslint-disable-next-line security/detect-unsafe-regex -- Star height 2
|
|
162
|
+
(`+` inside `(…)*`), but the inner group is prefixed by `/`, which is not
|
|
163
|
+
in [\w.-]. There is no ambiguous overlap, so matching stays linear. The
|
|
164
|
+
input is a short `owner/repo` slug that already failed the "://" and "@"
|
|
165
|
+
checks above. */
|
|
131
166
|
if (/^[\w.-]+\/[\w.-]+(\/[\w.-]+)*$/.test(trimmed)) {
|
|
132
167
|
return [`https://github.com/${trimmed}`, `git@github.com:${trimmed}.git`];
|
|
133
168
|
}
|
|
134
169
|
return [];
|
|
135
170
|
}
|
|
136
|
-
export async function lookupProjectByRepoName(repoName, host, token, verbose) {
|
|
171
|
+
export async function lookupProjectByRepoName(repoName, host, token, verbose, allowInsecureHttp = false) {
|
|
137
172
|
const candidates = repoNameToGitRemoteCandidates(repoName);
|
|
138
173
|
if (candidates.length === 0) {
|
|
139
174
|
if (verbose)
|
|
@@ -141,7 +176,7 @@ export async function lookupProjectByRepoName(repoName, host, token, verbose) {
|
|
|
141
176
|
return "not-found";
|
|
142
177
|
}
|
|
143
178
|
for (const candidate of candidates) {
|
|
144
|
-
const result = await lookupProjectByRemote(candidate, host, token, verbose);
|
|
179
|
+
const result = await lookupProjectByRemote(candidate, host, token, verbose, allowInsecureHttp);
|
|
145
180
|
if (result === "not-found")
|
|
146
181
|
continue;
|
|
147
182
|
return result;
|
|
@@ -164,7 +199,7 @@ export async function enrichCommitProjectAttribution(payloads, options) {
|
|
|
164
199
|
const repoName = payload.metadata?.repo_name;
|
|
165
200
|
if (!repoName)
|
|
166
201
|
continue;
|
|
167
|
-
const result = await lookupProjectByRepoName(repoName, options.host, options.token, options.verbose ?? false);
|
|
202
|
+
const result = await lookupProjectByRepoName(repoName, options.host, options.token, options.verbose ?? false, options.allowInsecureHttp === true);
|
|
168
203
|
if (result && typeof result === "object" && "project_id" in result) {
|
|
169
204
|
payload.project_id = result.project_id;
|
|
170
205
|
if (options.verbose) {
|
|
@@ -173,7 +208,18 @@ export async function enrichCommitProjectAttribution(payloads, options) {
|
|
|
173
208
|
}
|
|
174
209
|
}
|
|
175
210
|
}
|
|
176
|
-
export async function lookupProjectByRemote(gitRemote, host, token, verbose) {
|
|
211
|
+
export async function lookupProjectByRemote(gitRemote, host, token, verbose, allowInsecureHttp = false) {
|
|
212
|
+
const transportSecurity = evaluateTransportSecurity(host, {
|
|
213
|
+
allowInsecureHttp,
|
|
214
|
+
label: "DB90 project-lookup host",
|
|
215
|
+
});
|
|
216
|
+
if (!transportSecurity.ok) {
|
|
217
|
+
console.error(`Blocked project lookup — ${transportSecurity.error}`);
|
|
218
|
+
return null;
|
|
219
|
+
}
|
|
220
|
+
if (transportSecurity.warning) {
|
|
221
|
+
console.error(`Warning: ${transportSecurity.warning}`);
|
|
222
|
+
}
|
|
177
223
|
const url = `${host.replace(/\/$/, "")}/api/v1/projects/lookup?git_remote=${encodeURIComponent(gitRemote)}`;
|
|
178
224
|
try {
|
|
179
225
|
const res = await fetch(url, { headers: { Authorization: `Bearer ${token}` } });
|
|
@@ -0,0 +1,35 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Normalize an untrusted repo-path candidate. Pure — never touches the
|
|
3
|
+
* filesystem, so it is safe to call on every payload in a sync.
|
|
4
|
+
*
|
|
5
|
+
* Returns an absolute, `..`-collapsed path, or null when the value cannot be a
|
|
6
|
+
* legitimate workspace path. Rejecting relative values is deliberate: `git -C`
|
|
7
|
+
* would resolve a relative path against *this* process's cwd, which has nothing
|
|
8
|
+
* to do with where the value came from. It also rejects Cursor's literal
|
|
9
|
+
* `"unknown"` placeholder for global hook events.
|
|
10
|
+
*/
|
|
11
|
+
export declare function normalizeRepoPathCandidate(value: string | undefined | null): string | null;
|
|
12
|
+
/**
|
|
13
|
+
* True when `candidate` is `root` itself or lives beneath it.
|
|
14
|
+
*
|
|
15
|
+
* Compares with a trailing `sep` so `/repos/project-evil` is not treated as
|
|
16
|
+
* inside `/repos/project`, and resolves symlinks so a link inside the root
|
|
17
|
+
* cannot point out of it.
|
|
18
|
+
*
|
|
19
|
+
* When either side does not exist, `realpathSync` throws and the normalized
|
|
20
|
+
* paths are compared instead. That loses nothing — a path that does not exist
|
|
21
|
+
* cannot be a symlink, and `resolve()` has already collapsed `..` — and it keeps
|
|
22
|
+
* containment usable for scope filtering, which legitimately runs against
|
|
23
|
+
* payload paths naming directories this machine no longer has.
|
|
24
|
+
*/
|
|
25
|
+
export declare function isRepoPathWithinRoot(candidate: string, root: string): boolean;
|
|
26
|
+
/**
|
|
27
|
+
* The last check before `git -C <path>` runs. Requires the value to resolve to a
|
|
28
|
+
* real directory: a missing path, a dangling symlink, or a regular file is not a
|
|
29
|
+
* workspace. (Cursor's `metadata.workspace` is often the `state.vscdb` file
|
|
30
|
+
* itself, which git would only error on anyway.)
|
|
31
|
+
*
|
|
32
|
+
* Returns the canonical real path so `git` runs against exactly what was
|
|
33
|
+
* checked, narrowing the window between the check and the spawn.
|
|
34
|
+
*/
|
|
35
|
+
export declare function safeGitRepoPath(value: string | undefined | null): string | null;
|
|
@@ -0,0 +1,102 @@
|
|
|
1
|
+
import { realpathSync, statSync } from "node:fs";
|
|
2
|
+
import { isAbsolute, resolve, sep } from "node:path";
|
|
3
|
+
import { isSafeSpawnPathArg } from "./spawn-arg-safety.js";
|
|
4
|
+
/**
|
|
5
|
+
* Containment for untrusted filesystem paths that end up in `git -C <path>`.
|
|
6
|
+
*
|
|
7
|
+
* Every repo path this package resolves is untrusted text: Cursor's
|
|
8
|
+
* `workspace.json` `folder`, a composer's `workspaceIdentifier.uri.fsPath`, a
|
|
9
|
+
* hook's `workspace_roots[0]`, or a Claude transcript's `cwd`. None is validated
|
|
10
|
+
* by its producer — `fileUriToPath` (`readers/cursor.ts:189`) even passes a
|
|
11
|
+
* non-`file://` value straight through.
|
|
12
|
+
*
|
|
13
|
+
* `execFileSync` stops shell injection, but not `git -C ../../../elsewhere`: git
|
|
14
|
+
* would read that directory's `.git/config` and this package would ship the
|
|
15
|
+
* remote it found to the DB90 API. See DB90DV-547.
|
|
16
|
+
*
|
|
17
|
+
* Semantics are ported from `validatedRealPathWithinRoot`
|
|
18
|
+
* (`readers/cursor-sqlite.ts:23`), which already guards the Cursor SQLite
|
|
19
|
+
* reader the same way.
|
|
20
|
+
*/
|
|
21
|
+
function realPathOrNull(path) {
|
|
22
|
+
try {
|
|
23
|
+
return realpathSync(path);
|
|
24
|
+
}
|
|
25
|
+
catch {
|
|
26
|
+
return null;
|
|
27
|
+
}
|
|
28
|
+
}
|
|
29
|
+
/**
|
|
30
|
+
* Normalize an untrusted repo-path candidate. Pure — never touches the
|
|
31
|
+
* filesystem, so it is safe to call on every payload in a sync.
|
|
32
|
+
*
|
|
33
|
+
* Returns an absolute, `..`-collapsed path, or null when the value cannot be a
|
|
34
|
+
* legitimate workspace path. Rejecting relative values is deliberate: `git -C`
|
|
35
|
+
* would resolve a relative path against *this* process's cwd, which has nothing
|
|
36
|
+
* to do with where the value came from. It also rejects Cursor's literal
|
|
37
|
+
* `"unknown"` placeholder for global hook events.
|
|
38
|
+
*/
|
|
39
|
+
export function normalizeRepoPathCandidate(value) {
|
|
40
|
+
if (typeof value !== "string")
|
|
41
|
+
return null;
|
|
42
|
+
const trimmed = value.trim();
|
|
43
|
+
// Rejects empty, NUL-containing, and option-shaped values (DB90DV-546).
|
|
44
|
+
if (!isSafeSpawnPathArg(trimmed))
|
|
45
|
+
return null;
|
|
46
|
+
if (!isAbsolute(trimmed))
|
|
47
|
+
return null;
|
|
48
|
+
return resolve(trimmed);
|
|
49
|
+
}
|
|
50
|
+
/**
|
|
51
|
+
* True when `candidate` is `root` itself or lives beneath it.
|
|
52
|
+
*
|
|
53
|
+
* Compares with a trailing `sep` so `/repos/project-evil` is not treated as
|
|
54
|
+
* inside `/repos/project`, and resolves symlinks so a link inside the root
|
|
55
|
+
* cannot point out of it.
|
|
56
|
+
*
|
|
57
|
+
* When either side does not exist, `realpathSync` throws and the normalized
|
|
58
|
+
* paths are compared instead. That loses nothing — a path that does not exist
|
|
59
|
+
* cannot be a symlink, and `resolve()` has already collapsed `..` — and it keeps
|
|
60
|
+
* containment usable for scope filtering, which legitimately runs against
|
|
61
|
+
* payload paths naming directories this machine no longer has.
|
|
62
|
+
*/
|
|
63
|
+
export function isRepoPathWithinRoot(candidate, root) {
|
|
64
|
+
const normalizedCandidate = resolve(candidate);
|
|
65
|
+
const normalizedRoot = resolve(root);
|
|
66
|
+
const realCandidate = realPathOrNull(normalizedCandidate);
|
|
67
|
+
const realRoot = realPathOrNull(normalizedRoot);
|
|
68
|
+
// Compare like with like: mixing a realpath against a normalized path would
|
|
69
|
+
// false-negative on macOS, where /var is a symlink to /private/var.
|
|
70
|
+
const bothResolve = realCandidate !== null && realRoot !== null;
|
|
71
|
+
const left = bothResolve ? realCandidate : normalizedCandidate;
|
|
72
|
+
const right = bothResolve ? realRoot : normalizedRoot;
|
|
73
|
+
if (left === right)
|
|
74
|
+
return true;
|
|
75
|
+
const rootWithSep = right.endsWith(sep) ? right : `${right}${sep}`;
|
|
76
|
+
return left.startsWith(rootWithSep);
|
|
77
|
+
}
|
|
78
|
+
/**
|
|
79
|
+
* The last check before `git -C <path>` runs. Requires the value to resolve to a
|
|
80
|
+
* real directory: a missing path, a dangling symlink, or a regular file is not a
|
|
81
|
+
* workspace. (Cursor's `metadata.workspace` is often the `state.vscdb` file
|
|
82
|
+
* itself, which git would only error on anyway.)
|
|
83
|
+
*
|
|
84
|
+
* Returns the canonical real path so `git` runs against exactly what was
|
|
85
|
+
* checked, narrowing the window between the check and the spawn.
|
|
86
|
+
*/
|
|
87
|
+
export function safeGitRepoPath(value) {
|
|
88
|
+
const normalized = normalizeRepoPathCandidate(value);
|
|
89
|
+
if (normalized === null)
|
|
90
|
+
return null;
|
|
91
|
+
const real = realPathOrNull(normalized);
|
|
92
|
+
if (real === null)
|
|
93
|
+
return null;
|
|
94
|
+
try {
|
|
95
|
+
if (!statSync(real).isDirectory())
|
|
96
|
+
return null;
|
|
97
|
+
}
|
|
98
|
+
catch {
|
|
99
|
+
return null;
|
|
100
|
+
}
|
|
101
|
+
return real;
|
|
102
|
+
}
|
|
@@ -0,0 +1,25 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Guards for untrusted values that end up in the argv of a spawned process.
|
|
3
|
+
*
|
|
4
|
+
* `execFileSync` prevents *shell* injection but not *argv-option* injection: a
|
|
5
|
+
* value beginning with `-` is parsed by the child as a command-line option. Git
|
|
6
|
+
* remotes and workspace paths are untrusted text — they come from a repo the
|
|
7
|
+
* developer cloned, from Cursor's `workspace.json`, or from a Claude transcript
|
|
8
|
+
* — so every value derived from them must be checked before it reaches `git`
|
|
9
|
+
* or `ssh`. See DB90DV-546.
|
|
10
|
+
*/
|
|
11
|
+
/**
|
|
12
|
+
* True when `host` is safe to pass as an argv element to `ssh`. Accepts real
|
|
13
|
+
* hostnames, IPv4 literals, and `~/.ssh/config` host aliases.
|
|
14
|
+
*/
|
|
15
|
+
export declare function isSafeSshHost(host: string): boolean;
|
|
16
|
+
/**
|
|
17
|
+
* True when `value` is safe to pass as a filesystem-path argv element (e.g.
|
|
18
|
+
* after `git -C`). Deliberately permissive about path *content* — real
|
|
19
|
+
* workspace paths contain spaces, dashes and drive letters. It only rejects
|
|
20
|
+
* what makes the child misread the value as an option, plus embedded NUL.
|
|
21
|
+
*
|
|
22
|
+
* This is an argv guard, not a containment check: verifying the path points
|
|
23
|
+
* somewhere legitimate is DB90DV-547.
|
|
24
|
+
*/
|
|
25
|
+
export declare function isSafeSpawnPathArg(value: string): boolean;
|
|
@@ -0,0 +1,49 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Guards for untrusted values that end up in the argv of a spawned process.
|
|
3
|
+
*
|
|
4
|
+
* `execFileSync` prevents *shell* injection but not *argv-option* injection: a
|
|
5
|
+
* value beginning with `-` is parsed by the child as a command-line option. Git
|
|
6
|
+
* remotes and workspace paths are untrusted text — they come from a repo the
|
|
7
|
+
* developer cloned, from Cursor's `workspace.json`, or from a Claude transcript
|
|
8
|
+
* — so every value derived from them must be checked before it reaches `git`
|
|
9
|
+
* or `ssh`. See DB90DV-546.
|
|
10
|
+
*/
|
|
11
|
+
/** Longest legal DNS name (253) with headroom for an ssh_config alias. */
|
|
12
|
+
const MAX_HOST_LENGTH = 255;
|
|
13
|
+
/**
|
|
14
|
+
* Dot-separated labels of alphanumerics, `-` and `_`. A label may not start or
|
|
15
|
+
* end with `-`, which is what blocks option injection (`-oProxyCommand=…`).
|
|
16
|
+
* Whitespace, `=`, quotes, backslashes, newlines and NUL are all excluded.
|
|
17
|
+
* Underscores are allowed because `~/.ssh/config` aliases commonly use them.
|
|
18
|
+
* IPv6 literals are not covered — the SCP/`ssh://` host capture in
|
|
19
|
+
* `project-resolver.ts` cannot produce one, since it excludes `:`.
|
|
20
|
+
*/
|
|
21
|
+
const HOST_LABEL = "[A-Za-z0-9_](?:[A-Za-z0-9_-]*[A-Za-z0-9_])?";
|
|
22
|
+
const HOST_PATTERN = new RegExp(`^${HOST_LABEL}(?:\\.${HOST_LABEL})*$`);
|
|
23
|
+
/**
|
|
24
|
+
* True when `host` is safe to pass as an argv element to `ssh`. Accepts real
|
|
25
|
+
* hostnames, IPv4 literals, and `~/.ssh/config` host aliases.
|
|
26
|
+
*/
|
|
27
|
+
export function isSafeSshHost(host) {
|
|
28
|
+
if (host.length === 0 || host.length > MAX_HOST_LENGTH)
|
|
29
|
+
return false;
|
|
30
|
+
return HOST_PATTERN.test(host);
|
|
31
|
+
}
|
|
32
|
+
/**
|
|
33
|
+
* True when `value` is safe to pass as a filesystem-path argv element (e.g.
|
|
34
|
+
* after `git -C`). Deliberately permissive about path *content* — real
|
|
35
|
+
* workspace paths contain spaces, dashes and drive letters. It only rejects
|
|
36
|
+
* what makes the child misread the value as an option, plus embedded NUL.
|
|
37
|
+
*
|
|
38
|
+
* This is an argv guard, not a containment check: verifying the path points
|
|
39
|
+
* somewhere legitimate is DB90DV-547.
|
|
40
|
+
*/
|
|
41
|
+
export function isSafeSpawnPathArg(value) {
|
|
42
|
+
if (value.length === 0)
|
|
43
|
+
return false;
|
|
44
|
+
if (value.startsWith("-"))
|
|
45
|
+
return false;
|
|
46
|
+
if (value.includes("\0"))
|
|
47
|
+
return false;
|
|
48
|
+
return true;
|
|
49
|
+
}
|