# rag-loader-ssrf-v1

> rag-loader-ssrf-v1: 15 Keryn rules for python, typescript, javascript (12 error, 3 warning).

Source: https://keryn.dev/rules/rag-loader-ssrf-v1/

15 rules · Keryn Software Licence · rules 0.8.0 · [all packs](https://keryn.dev/rules/)

**How to read a rule**

- **Severity**: How bad the finding is if it is real: `error` fails a default gate, `warning` and `info` do not.

- **Confidence**: **P0** exploitable as written; **P1** dangerous in context; **P2** a heuristic or a posture check: confirm by reading the code.

- **False-positive risk**: **low**: an exact API or flag, so hits are almost always real; **medium**: the pattern is legitimate in many programs, so check each hit; **high**: it finds a lead to confirm, so expect benign hits.

- **Edge**: The relationship the finding sits on. **reaches**: content or traffic gets to a thing; **calls**: one thing invokes another; **reads**: one thing reads from another; **writes**: one thing writes to another; **sends**: data leaves for another party; **grants**: authority is given or left unchecked; **depends_on**: something relies on a component it does not verify; **trusts**: one party believes another without evidence. The boundary it crosses, when there is one: **user->prompt** (untrusted input reaching a prompt); **model->code** (model output reaching code, a shell, a query or a page); **secret->outside** (a secret leaving the trusted side); **supply-chain** (code, models or prompts pulled in from outside).

- **Taxonomy chips**: References for readers who work in OWASP LLM 2025, OWASP Agentic, CWE or NIST AI RMF. They say where a rule belongs, not that a framework is satisfied.

## seceng.ragloader.langchain-web-loader-ssrf

`error` python data-flow rule langchain, flask, fastapi

Request-controlled URL reaches a LangChain web loader (`WebBaseLoader`, `AsyncHtmlLoader`, `SeleniumURLLoader`). The server fetches it (and follows redirects), so a caller can reach cloud metadata (169.254.169.254), localhost admin ports and internal services, and the fetched text lands in your index. Validate scheme and host against an allow-list and block private/link-local addresses after DNS resolution.

Confidence **P1**: dangerous in context. False-positive risk **low**: an exact API or flag, so hits are almost always real.

Edge inspected: [reaches](https://keryn.dev/docs/edges/edge-kinds/#reaches) content or traffic gets to a thing; it crosses the [`user->prompt`](https://keryn.dev/docs/edges/trust-boundaries/#user-to-prompt) boundary (untrusted input reaching a prompt).

[CWE-918](https://keryn.dev/taxonomies/cwe/#cwe-918-server-side-request-forgery) [LLM01](https://keryn.dev/taxonomies/owasp-llm/#llm01-prompt-injection) [LLM02](https://keryn.dev/taxonomies/owasp-llm/#llm02-sensitive-information-disclosure) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack) [ASI02](https://keryn.dev/taxonomies/owasp-agentic/#asi02-tool-misuse-and-exploitation)

**Fix:** Parse the URL, require https and a hostname on an explicit allow-list, resolve it and reject loopback/private/link-local ranges (re-check after redirects), and fetch through an egress proxy. Do not let end users choose crawl roots.

**References:** [https://docs.langchain.com/oss/python/integrations/document_loaders](https://docs.langchain.com/oss/python/integrations/document_loaders)

**Try it: a snippet that triggers this rule**

Written for this page, not taken from any project. Save it as `rag/ingest_api.py` and scan it with `keryn scan rag --verbose`; the finding below appears.

```python
from flask import Flask, request, jsonify
from langchain_community.document_loaders import WebBaseLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter

app = Flask(__name__)
splitter = RecursiveCharacterTextSplitter(chunk_size=800, chunk_overlap=80)

@app.post("/ingest")
def ingest():
    page_url = request.form["url"]
    docs = WebBaseLoader(page_url).load()
    chunks = splitter.split_documents(docs)
    return jsonify(chunks=len(chunks))
```

Checked with Keryn's browser engine when this page was built: it reports `seceng.ragloader.langchain-web-loader-ssrf` at line 12 (error, proof 2). The rule's own logic is not published, so it cannot run in this tab; the pack list in the [playground](https://keryn.dev/playground/) shows what does.

## seceng.ragloader.recursive-url-loader-ssrf

`error` python data-flow rule langchain, flask, fastapi

Request-controlled URL reaches `RecursiveUrlLoader`, which also crawls every link it finds from that root. The server fetches it (and follows redirects), so a caller can reach cloud metadata (169.254.169.254), localhost admin ports and internal services, and the fetched text lands in your index. Validate scheme and host against an allow-list and block private/link-local addresses after DNS resolution.

Confidence **P1**: dangerous in context. False-positive risk **low**: an exact API or flag, so hits are almost always real.

Edge inspected: [reaches](https://keryn.dev/docs/edges/edge-kinds/#reaches) content or traffic gets to a thing; it crosses the [`user->prompt`](https://keryn.dev/docs/edges/trust-boundaries/#user-to-prompt) boundary (untrusted input reaching a prompt).

[CWE-918](https://keryn.dev/taxonomies/cwe/#cwe-918-server-side-request-forgery) [LLM01](https://keryn.dev/taxonomies/owasp-llm/#llm01-prompt-injection) [LLM02](https://keryn.dev/taxonomies/owasp-llm/#llm02-sensitive-information-disclosure) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack) [ASI02](https://keryn.dev/taxonomies/owasp-agentic/#asi02-tool-misuse-and-exploitation)

**Fix:** Parse the URL, require https and a hostname on an explicit allow-list, resolve it and reject loopback/private/link-local ranges (re-check after redirects), and fetch through an egress proxy. Do not let end users choose crawl roots.

**References:** [https://docs.langchain.com/oss/python/integrations/document_loaders](https://docs.langchain.com/oss/python/integrations/document_loaders)

## seceng.ragloader.unstructured-url-loader-ssrf

`error` python data-flow rule langchain, unstructured, flask, fastapi

Request-controlled URL reaches `UnstructuredURLLoader`. The server fetches it (and follows redirects), so a caller can reach cloud metadata (169.254.169.254), localhost admin ports and internal services, and the fetched text lands in your index. Validate scheme and host against an allow-list and block private/link-local addresses after DNS resolution.

Confidence **P1**: dangerous in context. False-positive risk **low**: an exact API or flag, so hits are almost always real.

Edge inspected: [reaches](https://keryn.dev/docs/edges/edge-kinds/#reaches) content or traffic gets to a thing; it crosses the [`user->prompt`](https://keryn.dev/docs/edges/trust-boundaries/#user-to-prompt) boundary (untrusted input reaching a prompt).

[CWE-918](https://keryn.dev/taxonomies/cwe/#cwe-918-server-side-request-forgery) [LLM01](https://keryn.dev/taxonomies/owasp-llm/#llm01-prompt-injection) [LLM02](https://keryn.dev/taxonomies/owasp-llm/#llm02-sensitive-information-disclosure) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack) [ASI02](https://keryn.dev/taxonomies/owasp-agentic/#asi02-tool-misuse-and-exploitation)

**Fix:** Parse the URL, require https and a hostname on an explicit allow-list, resolve it and reject loopback/private/link-local ranges (re-check after redirects), and fetch through an egress proxy. Do not let end users choose crawl roots.

**References:** [https://docs.langchain.com/oss/python/integrations/document_loaders](https://docs.langchain.com/oss/python/integrations/document_loaders)

## seceng.ragloader.playwright-url-loader-ssrf

`error` python data-flow rule langchain, playwright, flask, fastapi

Request-controlled URL reaches a browser-driven loader (`PlaywrightURLLoader`), which also executes the page's JavaScript. The server fetches it (and follows redirects), so a caller can reach cloud metadata (169.254.169.254), localhost admin ports and internal services, and the fetched text lands in your index. Validate scheme and host against an allow-list and block private/link-local addresses after DNS resolution.

Confidence **P1**: dangerous in context. False-positive risk **low**: an exact API or flag, so hits are almost always real.

Edge inspected: [reaches](https://keryn.dev/docs/edges/edge-kinds/#reaches) content or traffic gets to a thing; it crosses the [`user->prompt`](https://keryn.dev/docs/edges/trust-boundaries/#user-to-prompt) boundary (untrusted input reaching a prompt).

[CWE-918](https://keryn.dev/taxonomies/cwe/#cwe-918-server-side-request-forgery) [LLM01](https://keryn.dev/taxonomies/owasp-llm/#llm01-prompt-injection) [LLM02](https://keryn.dev/taxonomies/owasp-llm/#llm02-sensitive-information-disclosure) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack) [ASI02](https://keryn.dev/taxonomies/owasp-agentic/#asi02-tool-misuse-and-exploitation)

**Fix:** Parse the URL, require https and a hostname on an explicit allow-list, resolve it and reject loopback/private/link-local ranges (re-check after redirects), and fetch through an egress proxy. Do not let end users choose crawl roots.

**References:** [https://docs.langchain.com/oss/python/integrations/document_loaders](https://docs.langchain.com/oss/python/integrations/document_loaders)

## seceng.ragloader.llamaindex-web-reader-ssrf

`error` python data-flow rule llamaindex, flask, fastapi

Request-controlled URL reaches a LlamaIndex web reader (`SimpleWebPageReader`, `BeautifulSoupWebReader`, loaders from `download_loader`). The server fetches it (and follows redirects), so a caller can reach cloud metadata (169.254.169.254), localhost admin ports and internal services, and the fetched text lands in your index. Validate scheme and host against an allow-list and block private/link-local addresses after DNS resolution.

Confidence **P1**: dangerous in context. False-positive risk **low**: an exact API or flag, so hits are almost always real.

Edge inspected: [reaches](https://keryn.dev/docs/edges/edge-kinds/#reaches) content or traffic gets to a thing; it crosses the [`user->prompt`](https://keryn.dev/docs/edges/trust-boundaries/#user-to-prompt) boundary (untrusted input reaching a prompt).

[CWE-918](https://keryn.dev/taxonomies/cwe/#cwe-918-server-side-request-forgery) [LLM01](https://keryn.dev/taxonomies/owasp-llm/#llm01-prompt-injection) [LLM02](https://keryn.dev/taxonomies/owasp-llm/#llm02-sensitive-information-disclosure) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack) [ASI02](https://keryn.dev/taxonomies/owasp-agentic/#asi02-tool-misuse-and-exploitation)

**Fix:** Parse the URL, require https and a hostname on an explicit allow-list, resolve it and reject loopback/private/link-local ranges (re-check after redirects), and fetch through an egress proxy. Do not let end users choose crawl roots.

**References:** [https://developers.llamaindex.ai/python/framework-api-reference/readers/web/](https://developers.llamaindex.ai/python/framework-api-reference/readers/web/)

## seceng.ragloader.ingest-pipeline-fetch-user-url

`error` python data-flow rule rag, requests, httpx, flask, fastapi

Ingestion code fetches a request-controlled URL with an HTTP client and then chunks/embeds/indexes the response. Callers can make the server read internal endpoints (SSRF) and plant arbitrary text in your vector store. Allow-list hosts and block private ranges before fetching.

Confidence **P1**: dangerous in context. False-positive risk **medium**: the pattern is legitimate in many programs, so check each hit.

Edge inspected: [writes](https://keryn.dev/docs/edges/edge-kinds/#writes) one thing writes to another.

[CWE-918](https://keryn.dev/taxonomies/cwe/#cwe-918-server-side-request-forgery) [LLM01](https://keryn.dev/taxonomies/owasp-llm/#llm01-prompt-injection) [LLM08](https://keryn.dev/taxonomies/owasp-llm/#llm08-vector-and-embedding-weaknesses) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack) [ASI06](https://keryn.dev/taxonomies/owasp-agentic/#asi06-memory-and-context-poisoning)

**Fix:** Validate the URL (https, allow-listed hostname, resolved IP not private/loopback/link-local, redirects re-checked), fetch via an egress proxy with size and time limits, and record provenance for each indexed document.

**References:** [https://community.owasp.org/attacks/Server_Side_Request_Forgery](https://community.owasp.org/attacks/Server_Side_Request_Forgery)

## seceng.ragloader.langchain-file-loader-path-traversal

`error` python data-flow rule langchain, flask, fastapi

Request-controlled path reaches a LangChain file loader (`PyPDFLoader`, `TextLoader`, `UnstructuredFileLoader`, `CSVLoader`, ...). A caller can read arbitrary files via `../` or absolute paths and have them indexed and returned through the model. Resolve against a fixed base directory and reject anything that escapes it.

Confidence **P1**: dangerous in context. False-positive risk **low**: an exact API or flag, so hits are almost always real.

Edge inspected: [reads](https://keryn.dev/docs/edges/edge-kinds/#reads) one thing reads from another.

[CWE-22](https://keryn.dev/taxonomies/cwe/#cwe-22-path-traversal) [CWE-73](https://keryn.dev/taxonomies/cwe/#cwe-73-external-control-of-file-name-or-path) [LLM02](https://keryn.dev/taxonomies/owasp-llm/#llm02-sensitive-information-disclosure) [ASI02](https://keryn.dev/taxonomies/owasp-agentic/#asi02-tool-misuse-and-exploitation) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack)

**Fix:** Accept an identifier, not a path. Build the path with a fixed base (Path(BASE, name).resolve()), verify resolved.is\_relative\_to(BASE), use secure\_filename for uploads, and exclude symlinks.

**References:** [https://docs.langchain.com/oss/python/integrations/document_loaders](https://docs.langchain.com/oss/python/integrations/document_loaders)

## seceng.ragloader.directory-loader-path-traversal

`error` python data-flow rule langchain, flask, fastapi

Request-controlled path reaches `DirectoryLoader`, which globs and loads everything under that directory. A caller can read arbitrary files via `../` or absolute paths and have them indexed and returned through the model. Resolve against a fixed base directory and reject anything that escapes it.

Confidence **P1**: dangerous in context. False-positive risk **low**: an exact API or flag, so hits are almost always real.

Edge inspected: [reads](https://keryn.dev/docs/edges/edge-kinds/#reads) one thing reads from another.

[CWE-22](https://keryn.dev/taxonomies/cwe/#cwe-22-path-traversal) [CWE-73](https://keryn.dev/taxonomies/cwe/#cwe-73-external-control-of-file-name-or-path) [LLM02](https://keryn.dev/taxonomies/owasp-llm/#llm02-sensitive-information-disclosure) [ASI02](https://keryn.dev/taxonomies/owasp-agentic/#asi02-tool-misuse-and-exploitation) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack)

**Fix:** Accept an identifier, not a path. Build the path with a fixed base (Path(BASE, name).resolve()), verify resolved.is\_relative\_to(BASE), use secure\_filename for uploads, and exclude symlinks.

**References:** [https://docs.langchain.com/oss/python/integrations/document_loaders](https://docs.langchain.com/oss/python/integrations/document_loaders)

**Try it: a snippet that triggers this rule**

Written for this page, not taken from any project. Save it as `rag/folder_index.py` and scan it with `keryn scan rag --verbose`; the finding below appears.

```python
from flask import Flask, request, jsonify
from langchain_community.document_loaders import DirectoryLoader

app = Flask(__name__)

@app.get("/index-folder")
def index_folder():
    folder = request.args.get("folder", "docs")
    loader = DirectoryLoader(folder, glob="**/*.md")
    documents = loader.load()
    return jsonify(count=len(documents))
```

Checked with Keryn's browser engine when this page was built: it reports `seceng.ragloader.directory-loader-path-traversal` at line 10 (error, proof 2). The rule's own logic is not published, so it cannot run in this tab; the pack list in the [playground](https://keryn.dev/playground/) shows what does.

## seceng.ragloader.simple-directory-reader-path-traversal

`error` python data-flow rule llamaindex, flask, fastapi

Request-controlled path reaches LlamaIndex `SimpleDirectoryReader` (`input_dir` / `input_files`). A caller can read arbitrary files via `../` or absolute paths and have them indexed and returned through the model. Resolve against a fixed base directory and reject anything that escapes it.

Confidence **P1**: dangerous in context. False-positive risk **low**: an exact API or flag, so hits are almost always real.

Edge inspected: [reads](https://keryn.dev/docs/edges/edge-kinds/#reads) one thing reads from another.

[CWE-22](https://keryn.dev/taxonomies/cwe/#cwe-22-path-traversal) [CWE-73](https://keryn.dev/taxonomies/cwe/#cwe-73-external-control-of-file-name-or-path) [LLM02](https://keryn.dev/taxonomies/owasp-llm/#llm02-sensitive-information-disclosure) [ASI02](https://keryn.dev/taxonomies/owasp-agentic/#asi02-tool-misuse-and-exploitation) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack)

**Fix:** Accept an identifier, not a path. Build the path with a fixed base (Path(BASE, name).resolve()), verify resolved.is\_relative\_to(BASE), use secure\_filename for uploads, and exclude symlinks.

**References:** [https://developers.llamaindex.ai/python/framework/module_guides/loading/simpledirectoryreader/](https://developers.llamaindex.ai/python/framework/module_guides/loading/simpledirectoryreader/)

## seceng.ragloader.retrieved-content-in-system-prompt

`warning` python data-flow rule langchain, rag, openai, anthropic

Retrieved document text (`page_content`) is interpolated into the system/developer prompt with no delimiter or labeling. Any indexed page can then carry instructions with system-level authority (indirect prompt injection). Put retrieved text in a clearly delimited, labeled data block (or a user/tool message) and tell the model to treat it as untrusted.

Confidence **P2**: a heuristic or a posture check: confirm by reading the code. False-positive risk **high**: it finds a lead to confirm, so expect benign hits.

Edge inspected: [reaches](https://keryn.dev/docs/edges/edge-kinds/#reaches) content or traffic gets to a thing; it crosses the [`user->prompt`](https://keryn.dev/docs/edges/trust-boundaries/#user-to-prompt) boundary (untrusted input reaching a prompt).

[CWE-77](https://keryn.dev/taxonomies/cwe/#cwe-77-command-injection) [CWE-1427](https://keryn.dev/taxonomies/cwe/#cwe-1427-improper-neutralization-of-input-used-for-llm-prompting) [LLM01](https://keryn.dev/taxonomies/owasp-llm/#llm01-prompt-injection) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack) [ASI06](https://keryn.dev/taxonomies/owasp-agentic/#asi06-memory-and-context-poisoning)

**Fix:** Wrap retrieved chunks in explicit tags such as &lt;documents&gt;&lt;document source=...&gt;...&lt;/document&gt;&lt;/documents&gt;, state in the system prompt that their content is data and never instructions, strip or escape the closing tag inside chunks, and keep tool permissions independent of retrieved text.

**References:** [https://genai.owasp.org/llmrisk/llm01-prompt-injection/](https://genai.owasp.org/llmrisk/llm01-prompt-injection/) · [https://docs.langchain.com/oss/python/deepagents/rag](https://docs.langchain.com/oss/python/deepagents/rag)

## seceng.ragloader.loaded-doc-metadata-as-retrieval-filter

`warning` python data-flow rule langchain, chroma, pinecone, rag

Metadata taken from a loaded/retrieved document is used unchecked as a vector-store filter. Document metadata (source, author, tenant, tags) is attacker-influenced when documents come from the web or uploads, so it can widen retrieval across tenants or inject filter operators. Validate keys and values against a schema before filtering.

Confidence **P2**: a heuristic or a posture check: confirm by reading the code. False-positive risk **high**: it finds a lead to confirm, so expect benign hits.

Edge inspected: [calls](https://keryn.dev/docs/edges/edge-kinds/#calls) one thing invokes another; it crosses the [`model->code`](https://keryn.dev/docs/edges/trust-boundaries/#model-to-code) boundary (model output reaching code, a shell, a query or a page).

[CWE-943](https://keryn.dev/rules/?cwe=CWE-943) [CWE-20](https://keryn.dev/taxonomies/cwe/#cwe-20-improper-input-validation) [LLM08](https://keryn.dev/taxonomies/owasp-llm/#llm08-vector-and-embedding-weaknesses) [LLM01](https://keryn.dev/taxonomies/owasp-llm/#llm01-prompt-injection) [ASI06](https://keryn.dev/taxonomies/owasp-agentic/#asi06-memory-and-context-poisoning) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack)

**Fix:** Build filters from a server-side allow-list of keys and a typed schema (e.g. pydantic) and enforce tenant filters server-side; never copy metadata dicts from documents into filter arguments.

**References:** [https://docs.langchain.com/oss/python/langchain/overview](https://docs.langchain.com/oss/python/langchain/overview) · [https://docs.trychroma.com/docs/querying-collections/metadata-filtering](https://docs.trychroma.com/docs/querying-collections/metadata-filtering)

## seceng.ragloader.langchainjs-web-loader-ssrf

`error` typescript, javascript data-flow rule langchain, express, nextjs

Request-controlled URL reaches a LangChain.js web loader (`CheerioWebBaseLoader`, `PuppeteerWebBaseLoader`, `PlaywrightWebBaseLoader`, `FireCrawlLoader`). The server fetches it, so callers can reach cloud metadata and internal services. Validate scheme/host against an allow-list and block private ranges.

Confidence **P1**: dangerous in context. False-positive risk **low**: an exact API or flag, so hits are almost always real.

Edge inspected: [reaches](https://keryn.dev/docs/edges/edge-kinds/#reaches) content or traffic gets to a thing; it crosses the [`user->prompt`](https://keryn.dev/docs/edges/trust-boundaries/#user-to-prompt) boundary (untrusted input reaching a prompt).

[CWE-918](https://keryn.dev/taxonomies/cwe/#cwe-918-server-side-request-forgery) [LLM01](https://keryn.dev/taxonomies/owasp-llm/#llm01-prompt-injection) [LLM02](https://keryn.dev/taxonomies/owasp-llm/#llm02-sensitive-information-disclosure) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack) [ASI02](https://keryn.dev/taxonomies/owasp-agentic/#asi02-tool-misuse-and-exploitation)

**Fix:** Parse with new URL(), require https and an allow-listed hostname, resolve DNS and reject private/loopback/link-local addresses (re-check redirects), and crawl through an egress proxy.

**References:** [https://docs.langchain.com/oss/javascript/langchain/overview](https://docs.langchain.com/oss/javascript/langchain/overview)

**Try it: a snippet that triggers this rule**

Written for this page, not taken from any project. Save it as `src/ingest.js` and scan it with `keryn scan src --verbose`; the finding below appears.

```js
import express from "express";
import { CheerioWebBaseLoader } from "@langchain/community/document_loaders/web/cheerio";

const app = express();
app.use(express.json());

app.post("/ingest", async (req, res) => {
  const loader = new CheerioWebBaseLoader(req.body.url);
  const docs = await loader.load();
  res.json({ pages: docs.length, firstChars: docs[0]?.pageContent.slice(0, 120) });
});

app.listen(3000);
```

Checked with Keryn's browser engine when this page was built: it reports `seceng.ragloader.langchainjs-web-loader-ssrf` at line 8 (error, proof 2). The rule's own logic is not published, so it cannot run in this tab; the pack list in the [playground](https://keryn.dev/playground/) shows what does.

## seceng.ragloader.langchainjs-recursive-url-loader-ssrf

`error` typescript, javascript data-flow rule langchain, express, nextjs

Request-controlled URL is the crawl root for LangChain.js `RecursiveUrlLoader`, which follows every link it discovers. Callers can make the server crawl internal networks. Restrict roots to allow-listed hosts and cap depth.

Confidence **P1**: dangerous in context. False-positive risk **low**: an exact API or flag, so hits are almost always real.

Edge inspected: [reaches](https://keryn.dev/docs/edges/edge-kinds/#reaches) content or traffic gets to a thing; it crosses the [`user->prompt`](https://keryn.dev/docs/edges/trust-boundaries/#user-to-prompt) boundary (untrusted input reaching a prompt).

[CWE-918](https://keryn.dev/taxonomies/cwe/#cwe-918-server-side-request-forgery) [LLM01](https://keryn.dev/taxonomies/owasp-llm/#llm01-prompt-injection) [LLM02](https://keryn.dev/taxonomies/owasp-llm/#llm02-sensitive-information-disclosure) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack) [ASI02](https://keryn.dev/taxonomies/owasp-agentic/#asi02-tool-misuse-and-exploitation)

**Fix:** Allow-list crawl roots, set preventOutside: true and a small maxDepth, and block private address ranges after resolution.

**References:** [https://docs.langchain.com/oss/javascript/langchain/overview](https://docs.langchain.com/oss/javascript/langchain/overview)

## seceng.ragloader.langchainjs-file-loader-path-traversal

`error` typescript, javascript data-flow rule langchain, express, nextjs

Request-controlled path reaches a LangChain.js file loader (`PDFLoader`, `TextLoader`, `DirectoryLoader`, `CSVLoader`, ...). Callers can read arbitrary server files with `../` or absolute paths. Resolve under a fixed base directory and reject escapes.

Confidence **P1**: dangerous in context. False-positive risk **low**: an exact API or flag, so hits are almost always real.

Edge inspected: [reads](https://keryn.dev/docs/edges/edge-kinds/#reads) one thing reads from another.

[CWE-22](https://keryn.dev/taxonomies/cwe/#cwe-22-path-traversal) [CWE-73](https://keryn.dev/taxonomies/cwe/#cwe-73-external-control-of-file-name-or-path) [LLM02](https://keryn.dev/taxonomies/owasp-llm/#llm02-sensitive-information-disclosure) [ASI02](https://keryn.dev/taxonomies/owasp-agentic/#asi02-tool-misuse-and-exploitation) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack)

**Fix:** Take an id and map it to a file under a fixed base (path.resolve(BASE, name) then check startsWith(BASE + path.sep)), or use path.basename on upload names.

**References:** [https://docs.langchain.com/oss/javascript/langchain/overview](https://docs.langchain.com/oss/javascript/langchain/overview)

## seceng.ragloader.retrieved-content-in-system-prompt-ts

`warning` typescript, javascript data-flow rule langchain, rag, openai, anthropic

Retrieved document text (`pageContent`) is interpolated into the system/developer prompt with no delimiter or labeling, so any indexed page can carry instructions with system authority (indirect prompt injection). Put retrieved text in a delimited, labeled data block and tell the model it is untrusted.

Confidence **P2**: a heuristic or a posture check: confirm by reading the code. False-positive risk **high**: it finds a lead to confirm, so expect benign hits.

Edge inspected: [reaches](https://keryn.dev/docs/edges/edge-kinds/#reaches) content or traffic gets to a thing; it crosses the [`user->prompt`](https://keryn.dev/docs/edges/trust-boundaries/#user-to-prompt) boundary (untrusted input reaching a prompt).

[CWE-77](https://keryn.dev/taxonomies/cwe/#cwe-77-command-injection) [CWE-1427](https://keryn.dev/taxonomies/cwe/#cwe-1427-improper-neutralization-of-input-used-for-llm-prompting) [LLM01](https://keryn.dev/taxonomies/owasp-llm/#llm01-prompt-injection) [ASI01](https://keryn.dev/taxonomies/owasp-agentic/#asi01-agent-goal-hijack) [ASI06](https://keryn.dev/taxonomies/owasp-agentic/#asi06-memory-and-context-poisoning)

**Fix:** Wrap retrieved chunks in explicit tags such as &lt;documents&gt;...&lt;/documents&gt;, state in the system prompt that they are data and never instructions, escape closing tags inside chunks, and keep tool permissions independent of retrieved text.

**References:** [https://genai.owasp.org/llmrisk/llm01-prompt-injection/](https://genai.owasp.org/llmrisk/llm01-prompt-injection/) · [https://docs.langchain.com/oss/javascript/langchain/overview](https://docs.langchain.com/oss/javascript/langchain/overview)
