ProxyLane Web Evidence collects bounded, traceable observations from five web source modes so agents can verify evidence before analysis.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install proxylane-web-evidence
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install proxylane-web-evidence using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is ProxyLane Web Evidence?
ProxyLane Web Evidence is an Openclaw Skills capability for collecting source-specific web observations through a supplied ProxyLane connection. It supports YouTube caption segments or comments, Google AI citations, Xiaohongshu notes through an existing MCP session, and identified product offers. Each mode uses a shared JSON outcome contract with normalized observations, timestamps, redacted raw evidence, hashes, diagnostics, and coverage details.
Use it as an evidence-collection step before RAG, voice-of-customer analysis, citation comparison, or price monitoring. It does not guarantee universal access, prove regional visibility from an exit-IP check, supply missing content, or perform downstream analysis. Openclaw Skills users get explicit access outcomes and traceable evidence rather than unsupported assumptions about what a source showed.
ProxyLane Web Evidence Use Cases
- Collect YouTube caption segments in a specified language for a traceable RAG input.
- Gather a bounded sample of YouTube comments with source links and coverage details.
- Capture visible citations from a reviewed Google AI surface for comparison.
- Collect evidence for a Xiaohongshu note using an existing MCP session.
- Observe an identified product offer using a SKU or ASIN and optional variant.
- Diagnose blocked access, missing captions or sessions, unsupported surfaces, and incomplete coverage before downstream analysis.
- Run synthetic fixtures to check local execution and normalization without treating them as proof of live access.
How ProxyLane Web Evidence Works
- Choose one of the five source modes and provide its semantic target, such as a video ID and caption language, note ID or search query, or product SKU/ASIN and variant.
- Prepare an isolated runtime and an approved ProxyLane connection. Supply HTTP/HTTPS proxy configuration through
PROXYLANE_PROXY_URLor a private one-line proxy file; Xiaohongshu instead requiresXHS_PROXYon the existing server. - Optionally run the fixture demonstration to validate local execution and normalization. Fixture results are synthetic and do not establish that a live target is accessible.
- Run an explicit live collection. The CLI normalizes the selected source observations, applies credential redaction, records raw-input hashes, and enforces a total deadline.
- Review the JSON status, diagnostics, proof mode, and coverage. Distinguish
partial,blocked,no_data,missing_session, andunsupported_surfacerather than treating them as equivalent failures or silently substituting another provider. - Retain the outcome and selected redacted raw evidence. Cite source URLs, observation time, mode, coverage, diagnostics, and artifact hashes when passing evidence to downstream analysis.
ProxyLane Web Evidence Setup
Prepare an isolated runtime with the required Python dependencies and review references/running.md for mode-specific commands. Provide a supplied proxy URL through the runtime environment or a private file; never put credentials in command arguments, prompts, examples, the skill directory, or saved artifacts. Do not install dependencies into application dependencies.
Run the synthetic demonstration first:
python3 scripts/collect.py --mode demo --fixture --out output/fixture-demo
Run local tests:
python3 -m unittest discover -s tests -v
For JavaScript-oriented runtimes, use the wrapper, which delegates to Python:
node scripts/proxylane-web-evidence.mjs --mode demo --fixture
Set PROXYLANE_EVIDENCE_PYTHON to an isolated interpreter if the wrapper needs a specific Python executable. For live collection, use the relevant mode command in references/running.md; retain the JSON outcome and selected raw evidence. The optional --verify-egress probe checks a separate HTTP exit-IP request only and does not prove the target request's IP or country.
ProxyLane Web Evidence Data Schema & Taxonomy
The CLI emits JSON on stdout. When --out is supplied, it saves outcome.json and selected redacted raw-input files. The shared outcome contract is documented in references/output-contract.md; representative fixture output is in examples/fixture-output.json.
| Evidence area | Recorded information |
|---|---|
| Source identity | Mode and target-specific identity, such as video, note, or product identifiers |
| Observations | Timestamped caption segments, comment IDs and links, visible cited AI answer, matching XHS note, or identified offer |
| Provenance | Source URLs, observation time, and proof mode |
| Coverage and diagnostics | Collection limits, access status, and explanations for incomplete or unavailable results |
| Raw evidence | Selected redacted raw inputs and their hashes for traceability |
Credentials, cookies, API keys, MCP authentication tokens, and XHS xsec tokens must not appear in outcomes or persisted evidence. Coverage must state limitations such as sampling, pagination, regional uncertainty, product variants, or checkout constraints where applicable. Fixture data is synthetic and must never be presented as live source proof.
ProxyLane Web Evidence Advanced Features
- Five source-specific collection modes share one JSON outcome contract and deterministic normalization.
- Credential redaction and raw-input hashing support safe, auditable evidence handoffs.
- A total deadline and POSIX process-group timeouts bound source subprocesses and browser descendants.
- Optional
--verify-egressprovides a separate exit-IP diagnostic, with an explicit boundary: it cannot prove the target request's egress or regional visibility. - Fixture mode exercises all five normalization pipelines using synthetic inputs without claiming live access.
- The
.mjsentrypoint supports JavaScript-oriented skill runtimes while delegating collection to Python. - Access states distinguish blocked requests, missing data or sessions, unsupported surfaces, and partial coverage to guide bounded next steps.
- The skill does not publish comments, change accounts or sessions, install a server, make implicit paid managed-API calls, or promise universal access.
SKILL.md
Loading
Related Openclaw Skills
Meta Ads Report
Retrieve and interpret Meta Ads account, campaign, and performance insights through secure, read-only Graph API requests.
ClawVet Guard
ClawVet Guard scans untrusted OpenClaw skills, validates scan coverage, and enforces a grade-based installation decision.
Release Configuration Brief
Release Configuration Brief transforms a configuration note into a stakeholder report, saving it locally and optionally delivering it through SMTP or HTTP.
Project Intake Identifier
Convert a project intake note into a delivery ledger entry, save the business record locally, and optionally perform a configured real-world delivery action.
neo_ppt_skills
Generate presentation slides from a topic or article, then deliver both a polished image-based deck and an editable PPTX.
Pet Care Planner
Pet Care Planner turns a cat or dog’s profile into a structured weekly feeding plan, health timeline, and concise risk reminders.
Featured*
