registry / AI-agent capability abuse / hippo-memory / 1.52.9

hippo-memory@1.52.9

Memory for AI agents that learns what is wrong and ranks it down. MCP server, hooks for Claude Code, OpenCode and Codex, AGENTS.md instructions for Codex, Cursor, OpenClaw and Pi. SQLite, zero runtime deps.

AI Security Review

scanned 15d ago · by lpm-firewall-ai

The source shows an agent configuration and scheduling capability activated by the explicit `hippo init` command, which warrants a warning under the agent integration policy. The inspected postinstall path does not establish unconsented first-time setup, and the excerpts show no concrete malicious behavior.

Static reason

One or more suspicious static signals were detected.

Trigger

A user runs `hippo init`; the package postinstall only repairs a previously installed Codex wrapper and prints setup guidance.

Impact

Agent configuration and scheduled local maintenance may be added after the explicit CLI command. No confirmed exfiltration or remote execution was identified in the inspected excerpts.

Mechanism

The CLI detects agent marker files and installs hooks or instructions; the init flow can also configure a daily runner task.

AI rationale

Evidence

dist/cli.jsdist/scheduler.jsdist/postinstall.jsdist/capture.jsdist/hooks.jsdist/hooks.d.ts

Decision evidence

public snapshot

AI called this Suspicious at 84.0% confidence as Dangerous Capability with medium false-positive risk.

Evidence for warning

package.json declares a postinstall script and describes the package as providing agent hooks and instructions.
The postinstall repair path is limited to a Codex wrapper already installed by the user; first-time setup is offered as a CLI command.
The `hippo init` flow can install hooks and set up a daily schedule when the user runs it.
The CLI detects existing agent marker files, including Claude Code settings and Codex instruction files.
The capture helper reads Claude project transcript directories; the cited hook declaration exposes OpenCode plugin install and uninstall functions, while the cited hook helper checks for Codex configuration.
The scheduler builds a daily task command for the Hippo runner.

Evidence against

The postinstall source describes its Codex wrapper behavior as repair-only and presents first-time setup as an explicit command.
The inspected capture and scheduler excerpts show local file or task behavior; they do not establish credential exfiltration or a remote payload.

Behavioral surface

Source

ChildProcess

Source & flagged code

8 flagged

package.jsonView file

Lines 19-59json

19 "files": [

20 "dist",

21 "!dist/src",

22 "!dist/**/*.map",

23 "dist-ui",

24 "bin",

25 "scripts/postinstall.cjs",

26 "openclaw.plugin.json",

27 "extensions/openclaw-plugin"

28 ],

29 "scripts": {

30 "baseline": "npm run build && npm test",

31 "build": "node -e \"require('fs').rmSync('dist',{recursive:true,force:true})\" && tsc && tsc -p tsconfig.benchmarks.json && tsc -p tsconfig.extensions.json",

32 "dev": "tsc --watch",

33 "pretest": "npm run build",

34 "test": "vitest run",

35 "test:watch": "vitest",

36 "lint": "oxlint",

37 "sbom": "node scripts/sbom.mjs",

38 "audit:security": "npm audit --audit-level=high && npm --prefix ui audit --audit-level=high",

39 "postinstall": "node scripts/postinstall.cjs",

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg · L39

Medium

Ambiguous Install Lifecycle Script

Install-time lifecycle script is not statically allowlisted and needs review.

package.jsonView on unpkg · L39

40 "smoke:pack": "node scripts/smoke-pack.mjs",

41 "smoke:openclaw-install": "node scripts/smoke-openclaw-install.mjs",

42 "build:ui": "cd ui && npm install && npm run build",

43 "build:all": "npm run build && npm run build:ui",

44 "version": "node scripts/sync-version.mjs && node scripts/check-manifest-versions.mjs",

45 "prepublishOnly": "node scripts/check-manifest-versions.mjs && node scripts/changelog-fragments.mjs check && node scripts/check-em-dashes-in-release-notes.mjs && node scripts/check-graph-writes.mjs && npm run build:all && node scripts/check-openclaw-dist.mjs && npm run smoke:openclaw-install && node scripts/check-tests-pass.mjs"

46 },

47 "keywords": [

48 "memory",

49 "agent-memory",

50 "ai-memory",

51 "llm-memory",

52 "long-term-memory",

53 "persistent-memory",

54 "memory-layer",

55 "mcp",

56 "mcp-server",

57 "model-context-protocol",

58 "claude-code",

59 "claude",

dist/embeddings.jsView file

Lines 80-120javascript

80 * official inference code in `FlagEmbedding`); MiniLM and most sentence-

81 * transformers models use mean pooling. Unknown model ids default to mean

82 * — that is the safe choice because most third-party models adopt the

83 * sentence-transformers convention, and the

dist/rerankers/cross-encoder.jsView file

Lines 1-23javascript

1import { createRequire } from 'node:module';

2import { pathToFileURL } from 'node:url';

3const MODEL_NAME = 'Xenova/ms-marco-MiniLM-L-6-v2';

4const _require

dist/scheduler.jsView file

Lines 10-50javascript

10/** Test-only override. Not part of this module's public API surface. */

11export function __setSchedulerFsDeps(overrides) {

12 fsDeps = { ...fsDeps, ...overrides };

13}

14function defaultRegistry() {

dist/cli.jsView file

•Manifest-trigger-reachable source links an external AI-agent control path to a behavior-bearing write operation. console.error(`unknown hook target: ${target ?? '(none)'}`); process.exit(1); } const hook = hooks[target]; const filepath = path.resolve(process.cwd(), hook.file); if (fs.existssync(filepath)) { const existing = fs.readfilesync(filepath, 'utf8'); if (existing.includes(hook_markers.start) && othersblock(existing)) { const owner = hippoblock(existing)?.owner; const whose = owner ? `hippo wrote it for ${owner}` : 'it has been edited, so hippo cannot tell whose it is'; console.log(`left the hippo block in ${hook.file}: ${whose}. delete it by hand if no agent needs it.

High

Trigger Reachable External Ai Agent Control Surface Mutation

Manifest-trigger-reachable source writes behavior-bearing configuration into a user or project AI-agent control surface.

dist/cli.jsView on unpkg

dist/embedding-provider.js#virtual:normalized:round1View file

•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 5

High

Semantic Analysis Limited

A bounded semantic-analysis stage reached its safety limit; remaining detectors completed, but this package requires AI review.

dist/embedding-provider.js#virtual:normalized:round1View on unpkg

Source & flagged code

8 flagged · loading source

package.jsonView file

•scripts.postinstall = node scripts/postinstall.cjs

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg

•scripts.postinstall = node scripts/postinstall.cjs

Medium

Ambiguous Install Lifecycle Script

Install-time lifecycle script is not statically allowlisted and needs review.

package.jsonView on unpkg

dist/embeddings.jsView file

100// Promise<object> names that honestly without claiming a specific module shape. L101: const _dynImport = new Function('s', 'return import(s)'); L102: /**

Low

Eval

Package source references a known benign dynamic code generation pattern.

dist/embeddings.jsView on unpkg · L100

Findings

4 High6 Medium6 Low

HighInstall Time Lifecycle Scriptspackage.json

HighTrigger Reachable External Ai Agent Control Surface Mutationdist/cli.js

HighTrigger Reachable Persistencedist/scheduler.js

HighSemantic Analysis Limiteddist/embedding-provider.js#virtual:normalized:round1

MediumAmbiguous Install Lifecycle Scriptpackage.json

MediumDynamic Requiredist/rerankers/cross-encoder.js

MediumNetwork

MediumEnvironment Vars

MediumInstall Persistencedist/scheduler.js

MediumStructural Risk Force Deep Review

LowNon Install Lifecycle Scripts

LowScripts Present

LowEvaldist/embeddings.js

Affected versions and remediation

This report applies to hippo-memory@1.52.9.

See all versions of hippo-memory for other recorded verdicts.

Review the evidence and your use of hippo-memory@1.52.9 before allowing it. Restrict the permissions described in this report.
Choose an independently verified alternative or release. This report does not establish that other versions are safe.

Evidence last updated: 2026-10-07 20:27:03.861Z (UTC).

Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Affected versions and remediation

Runtime surface

Behavioral surface

Source & flagged code