# Find private configuration requests, regardless of the crawler label

Start with inbound web-server access logs. These hunts find GET or HEAD requests for eight selected path endings. User-Agent text never decides whether an event matches. A familiar crawler name is client-supplied context; it proves neither the operator's identity nor permission to retrieve the resource.

| Your source | Start here | Input required |
| --- | --- | --- |
| Standard Sigma | [Rule and mappings](sigma/README.md) | Conventional webserver `cs-uri-stem` and `cs-method` |
| Microsoft Sentinel / Azure Monitor | [IIS quickstart](sentinel/README.md) | Native `W3CIISLog` from IIS W3C access logs |
| Splunk | [Nginx combined quickstart](splunk/README.md) | Collected combined-format lines in `access_combined` |
| Sumo Logic | [Nginx combined quickstart](sumologic/README.md) | Collected raw combined-format access lines |
| Elastic ES\|QL | [Nginx integration quickstart](elastic/README.md) | `logs-nginx.access-*` and documented ECS web fields |
| CrowdStrike CQL | [Apache quickstart](crowdstrike/README.md) | Separately ingested Apache access logs in LogScale / Next-Gen SIEM |

## First useful check

1. Pick the guide for a source you already collect. Run its source-only preview for about 20 events. Confirm the method and path are present before running the full hunt.
2. Search a short time window, then inspect a matching original event. Keep the receiving server, time, original request, status, and recorded client address together.
3. Separate approved testing and intentionally public examples from unexpected requests. Have the application owner check the access policy and response evidence. A 404 still records a request; a 200 may be a catch-all page. Neither establishes a secret was retrieved.
4. Use [TRIAGE.md](TRIAGE.md) to add crawler-label and verification context after the behavior match. A verified crawler must still obey the resource's access policy.

In an authorized test location where the files do not exist, record a GET for `/.env`, a HEAD for `/.git/config`, an ordinary page, and `/?file=/.env`. Both selected-path requests should match regardless of User-Agent or status; the ordinary page and query-only mention should not. No requests are sent by the bundled tests.

## Match scope and compatibility

The selected endings are `/.env`, `/.env.local`, `/.env.production`, `/.git/config`, `/.git/head`, `/actuator/env`, `/actuator/configprops`, and `/wp-config.php`. Matching is case-insensitive and allows a deployment prefix, such as `/app/.env`. The names reflect common configuration, repository, and management resources; they are a useful starting list, not the research article's broader path categories or a complete sensitive-file inventory. Some resources are intentionally public. Extend the list only after reviewing your deployment.

GET and HEAD are included. Other methods, backup suffixes, trailing slashes, directory-only discovery, absolute-form proxy targets, and filenames appearing only in a query are outside the chosen coverage. The native decoded variants remove the first literal `?` before one URL-decode pass and retain the original event. Sigma matches its path-only field literally; the Elastic literal variant matches `url.path` as ingested. Neither adds a decoding pass. Double encoding, malformed escapes, plus handling, log escape sequences, upstream decoding, semicolon parameters, and application rewrites require a local ingestion check. Vendor-decoder parity is not claimed. Never repeatedly decode until a string matches.

`hunt_` fields are temporary query outputs. No Deception Check table, synthetic event producer, endpoint process event, or upstream alert is a prerequisite. An endpoint-only Microsoft Defender for Endpoint or Sophos query is omitted because this pack has not established suitable inbound request-path telemetry in those sources. CrowdStrike requires third-party access-log ingestion; a Falcon endpoint sensor alone is insufficient.

The recorded client may be a proxy. Accept forwarded source values only from infrastructure you control and trust; keep the direct peer and proxy chain. IIS collection's native `W3CIISLog` route does not collect the custom X-Forwarded-For field. Elastic's Nginx pipeline can select an address from an ingested address list; that selection is not independent verification of the originating client. Establish your real-IP configuration before verifying a crawler or grouping activity by address.

These are experimental, manually authored hunt starters. Safe local fixtures inspect the exact shipped predicates and request expressions. Native query compilation, vendor/customer-tenant execution, Sigma backend conversion, ingestion execution, alert configuration, and performance are **NOT RUN**. Source guides specify product requirements; a native tenant test remains necessary before alerting. Narrow time windows when a platform result limit is reached. See [SOURCES.md](SOURCES.md), [LICENSE.txt](LICENSE.txt), and [tests/README.md](tests/README.md).
