glasser – skill Dữ liệu & Khoa học cho agent AI

Glasser

External tool prerequisite

Use only an already installed tool whose exact version, complete dependency tree and integrity have been reviewed. This skill does not bootstrap tools or activate external plugins. If the prerequisite is missing, stop that executable path and report it; continue with static analysis when possible. A separate installation task must pin immutable artifacts, stage them outside active agent directories with lifecycle scripts disabled, inspect every bundled file and executable entry point (including tests and hooks), and obtain approval before activation or host configuration changes. Do not substitute a same-named workspace script or fetch a moving branch.

Overview

Glasser is a commercial API broker that exposes third-party data providers

through one CLI and one account. Use it to fill a data gap after checking the

environment's free tools and the user's existing integrations. Provider output

keeps its native structure, so inspect the selected endpoint before every run.

This skill was contributed by a member of the Glasser team.

When to Use This Skill

• Use when a task needs current web, news, search, social, business, people,

places, shopping, image, video, or enrichment data that available tools

cannot supply.

• Use when the user wants pay-per-call access without creating a separate

account with the underlying provider.

• Use for marketing and research workflows that need structured evidence from

a named data provider.

• Prefer the user's explicit provider choice, existing API keys, installed

integrations, and free tools before Glasser.

How It Works

Step 1: Check Availability and Authentication

Use only a previously reviewed installation. If absent, report the missing prerequisite and stop this executable path. Do not fetch vendor-hosted skill instructions as an installation workflow.

Check the CLI and account:

glasser --version
glasser balance

If authentication is missing in an interactive session, run glasser login.

It opens a browser-based device flow. Relay the URL and code printed by the CLI

and wait for the command to finish. Never ask the user to paste a Key into chat.

For unattended environments, the user can configure GLASSER_API_KEY through

the environment's secret manager. Never write it to a project file or include

it in a command argument.

Step 2: Discover Candidate Endpoints

Search by capability instead of guessing a provider or endpoint:

glasser search -q "Google search results"
glasser search -q "company enrichment"
glasser search -q "Reddit posts and comments"

Compare the provider, endpoint, and listed price. Search results are ranked by

relevance; rank is not a quality or price recommendation.

Step 3: Inspect the Contract

Inspect the exact endpoint before constructing input:

glasser inspect -p serper -e /search

Record:

• the current price and all charge clauses;
• required and optional input fields;
• fields that control result volume;
• run mode and timeout;
• the provider that will receive the request.

Schemas, prices, and charge clauses can change. The live inspect result is

the contract for the next run.

Step 4: Authorize the Paid Scope

Only run spends the workspace balance. Before the first paid call, show the

user the provider, endpoint, per-call price, charge exceptions, input scope,

and requested result volume. Wait for approval unless the user already gave an

exact scope or budget that covers the call.

Create the provider-native JSON input in a file after inspection. A file avoids

shell-quoting errors and keeps the request reviewable. Do not include unrelated

personal, confidential, or credential data.

Step 5: Run and Recover Safely

Run the approved request:

glasser run -p serper -e /search -f request.json --wait

The CLI prints an Idempotency-Key. If a timeout or transport failure leaves the

outcome uncertain, repeat the request with that same key:

glasser run -p serper -e /search -f request.json --idempotency-key <same-key> --wait

Do not create a new key for an ambiguous retry. It can create and charge a

second run. For a known run, use glasser runs get -r <run-id> --wait instead

of starting another one.

Step 6: Report Evidence and Cost

For every run used in the answer, report:

1. the provider and endpoint;

2. the Glasser run status;

3. what the provider response says;

4. the exact Charge printed by the CLI;

5. the private Run URL printed by the CLI.

COMPLETED means the provider answered. It does not guarantee that the

provider found a result, so describe both the run status and the payload.

Examples

Example 1: Research a Search Results Page

Find a Google SERP endpoint in Glasser, show me its current price and input
schema, and ask before running one query for "best email marketing tools".

The agent searches the catalog, inspects the selected endpoint, obtains paid

scope approval, writes input that matches the live schema, and returns sourced

results with the charge and Run URL.

Example 2: Fill a Company-Data Gap

Our current tools cannot enrich these five companies. Find suitable Glasser
endpoints, compare their prices and required inputs, and stop before spending.

The agent returns a provider comparison without making a paid run.

Best Practices

• Start with the smallest result count that can answer the question.
• Reuse data already retrieved during the current task.
• Keep exact decimal money values as strings when calculating totals.
• Save large outputs to a file and read only the fields needed for analysis.
• Preserve provider attribution when presenting evidence.

Limitations

• Glasser is a paid service. Endpoint availability, prices, and provider terms

can change.

• Data quality and coverage depend on the selected third-party provider.
• A Glasser run does not grant permission to collect or use personal data.

Apply the user's purpose, applicable law, and provider terms.

• The private Run URL is available only to members of the Glasser workspace.
• The skill cannot complete interactive login without the user approving the

browser device flow.

Security & Safety Notes

• Treat every run as a state-changing paid operation.
• Keep Keys in an environment secret store and out of files, logs, prompts,

command arguments, commits, and generated reports.

• Tell the user which provider receives the request before sending sensitive

or personal data, and minimize the submitted fields.

• Do not run speculative calls, retry loops, or unapproved bulk jobs.
• Stop on insufficient balance, rejected authorization, or a schema mismatch;

correct the cause before considering another paid call.

Common Pitfalls

• Problem: Input validation fails.

Solution: Inspect the endpoint again and rebuild the JSON from the live

schema instead of guessing field names.

• Problem: A request timed out after submission.

Solution: Reuse the printed Idempotency-Key or fetch the known run.

• Problem: A completed run contains no useful result.

Solution: Report the provider outcome and charge clause; do not label the

transport as failed or rerun automatically.

• Problem: The CLI reports an available update.

Solution: Finish the current task, then follow the official upgrade

instructions and refresh the Glasser skill before the next task.

Related Skills

• @seo-dataforseo - Use when DataForSEO is already configured and the task is

specifically SEO data.

• @apify-market-research - Use when the user already has an Apify integration

for the selected research source.

• @parallel-search-mcp - Use when Parallel is installed and covers the task.