Webresearch

/webresearch

Effectiveness-first current web research for APIs, versions, docs, errors, pricing, standards, and other time-sensitive facts.

MiniMax MCP is the primary research tool. Prefer mcp__MiniMax__web_search whenever it is available.

MAXPARALLELAGENTS — ceiling for web research lanes. Use only the number of distinct questions that materially change the answer.

Use when: The user wants the latest docs, current behavior, recent changes, known issues, exact error explanations, or source-backed technical guidance from the web.

Claude product questions: Route Claude, Claude Code, Claude.ai, Anthropic API, connector, plugin, skill, hook, MCP, subagent, plan availability, limit, or setup questions through /claudeproduct first. It constrains this research mode to official Anthropic/Claude sources and separates product surfaces before answering.

Research-First: Integrated into /workflow automatically. Use /webresearch directly when the main task is current-fact verification.

Swarm: "swarm webresearch" → /webresearch with an effectiveness-first research wave up to MAX_PARALLEL_AGENTS.

If the task becomes multi-branch, strategic, adversarial, or investigation-heavy, escalate to /deepresearch.

---

Contract

/webresearch is the focused current-facts sibling of /deepresearch.

It still requires:

• the current minmaxing time anchor for relative-date or current-fact claims
• a concise collaborative research plan
• an effective budget up to MAX_PARALLEL_AGENTS
• at least one search -> read -> refine cycle when external facts matter
• a source ledger when the result will drive implementation or decision-making

Temporal guard:

• Use the minmaxing temporal anchor injected by hooks, or run

bash scripts/time-anchor.sh text, before answering current-fact questions.

• For "latest", "today", "current", "recent", "SOTA 2026", model/provider

behavior, pricing, docs, laws, standards, benchmarks, schedules, or news, cite live sources and include source publish/update dates plus access date.

• If current verification cannot be completed, say so explicitly instead of

filling gaps from pretrained memory.

When To Escalate To /deepresearch

Escalate when:

• more than 3 branches materially affect the answer
• the user explicitly asks for "deep research" or high-confidence investigation
• conflicting evidence appears and the stakes are non-trivial
• the answer will drive architecture, security, or substantial implementation time

Step 1: Memory Recall

bash scripts/memory.sh recall "[research topic]" --depth simple 2>/dev/null || echo "Memory recall: skipped"
bash scripts/memory.sh search "[topic]" 2>/dev/null || true

Step 2: Collaborative Research Plan

Define:

• the exact question
• what current facts are needed
• what source classes count as trustworthy
• what would make the answer sufficient

Step 3: Effective Budget

Typical lanes:

• official docs
• release notes / changelog
• GitHub issues / discussions
• recent practitioner writeups

Use the smallest effective budget:

effective_webresearch_budget = min(MAX_PARALLEL_AGENTS, distinct_questions, reviewer_capacity)

Step 4: Search -> Read -> Refine

1. discovery wave
2. open the highest-value sources
3. run one follow-up wave if the first pass leaves decision-relevant gaps

Step 5: Source Ledger

When the result materially affects implementation, record:

• cited
• reviewed but not cited
• rejected / downweighted

Output

## WebResearch: [Topic]

### Collaborative Research Plan
- Question: ...
- Needed facts: ...
- Stop Condition: ...

### Research Tracks
| Track | Query | Sources |
|-------|-------|---------|

### Findings
- ...

### Source Ledger
- Cited: ...
- Reviewed but not cited: ...
- Time Anchor: ...
- Access Date: ...

### Coverage
- Research Tracks Used: [completed] / [effective budget]
- MiniMax MCP Searches: [count]
- Follow-up Research: [not needed / completed / escalated to /deepresearch]

Quality Gates

• cite the sources
• prioritize official docs for technical claims
• keep the budget effectiveness-first
• escalate to /deepresearch when the problem is no longer narrow

Anti-Patterns

• answering time-sensitive questions from memory
• using all lanes just because MAX_PARALLEL_AGENTS allows it
• pretending a narrow docs lookup was a full deep investigation

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: waitdeadai
• Source: waitdeadai/minmaxing
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.