Ai Research Eta Optimization

Optimizes AI research ETAs with dynamic updates, parallel execution, smart filtering, and template-driven workflows to accelerate prospect analysis by 2-5x w...

Security Scan

This is an instruction-only workflow for speeding up AI research ETAs; its instructions, scope, and lack of required credentials/install steps are coherent with its stated purpose.

Purpose & Capability

The name/description (ETA optimization for AI research) matches the SKILL.md content: protocols for initial estimates, parallelization, filtering, templates, and communication. There are no unexpected binaries, config paths, or environment variables requested that would be unrelated to this purpose.

Instruction Scope

Instructions direct concurrent web_search() calls, verification against public sources (LinkedIn, org charts, SEC filings, 'dark web' claims) and to send stakeholder progress updates. Those actions are consistent with prospect research, but the SKILL.md leaves communication methods and external endpoints unspecified—giving an agent discretion about how updates are sent and what external services/APIs are used.

Install Mechanism

No install spec and no code files — the skill is instruction-only, so nothing is written to disk or pulled from external URLs during installation.

Credentials

The skill requests no environment variables or credentials, which is appropriate for a guidance-only skill. However, it references data sources (LinkedIn, SEC filings, other high-signal sources) that in practice may require API keys or access methods; the SKILL.md does not declare or justify any such credentials.

Persistence & Privilege

The skill does not request persistent presence (always:false), does not modify other skills or system configs, and requires no privileged access.

Assessment

This skill appears coherent and lightweight, but before installing consider: (1) confirm how your agent will perform web_search() and what connectors it will use (some sources need API keys or paid access); (2) decide and lock down how progress updates/notifications are delivered (email, Slack, etc.) to avoid accidental data leaks; (3) be cautious about scraping or collecting personal data (LinkedIn, org charts) and ensure you comply with privacy/legal rules; (4) test the workflow on non-sensitive sample tasks to verify it doesn't attempt to access undeclared services or credentials. If you require the skill to use specific APIs, plan to provide those credentials through your normal secure channels rather than expecting the skill to create them.

AI Research Task ETA Optimization Workflow

Created: 2026-04-11
Based on: Arkose Labs 50-Prospect Research (1h 41m vs. 5-6h estimate)
Status: ✅ Operational

---

Quick Reference Formula

AI Research Time = Human Benchmark × 0.2-0.3
Standard Estimate - 30% buffer = Realistic AI Timeline
Optimized scenarios = Standard Estimate - 50%

Example:

• Human research (50 prospects): 8-10 hours
• AI agent optimized: 2-3 hours
• With parallel execution: 1-2 hours

---

Dynamic ETA Protocol

Hour 1 (Start): Conservative Initial Estimate

• Commit to conservative timeline
• Add note: "ETA may accelerate based on execution patterns"
• Do not commit to fixed deadline

Hour 2 (Mid-Task): Re-evaluate

• Check if accelerating or decelerating
• Look for parallel execution opportunities
• Adjust ETA if needed
• If accelerating: Notify stakeholder early

Hour 3+: Confirm or Finalize

• Confirm final ETA
• Adjust if unexpected patterns emerge
• Document acceleration/deceleration factors

---

Parallel Execution Optimization

When to Parallelize

• Multiple web searches
• Cross-source verification
• Data aggregation
• Template filling

How to Parallelize

✅ Use concurrent web_search() calls
✅ Batch data verification tasks
✅ Run multiple source queries simultaneously
✅ Avoid sequential bottlenecks

Impact: +30-60 minutes on 2-4 hour tasks

---

Smart Filtering Framework

Early Elimination Criteria

• Low fraud signal: <2 verifiable incidents
• Weak Arkose fit: No clear value proposition
• Missing decision-makers: Cannot identify contacts
• Low urgency: No recent incidents or regulatory pressure

Prioritization Strategy

1. Tier 1 (Urgent): Recent breach + regulatory action + clear value prop
2. Tier 2 (High): Strong fraud signal + good fit
3. Tier 3 (Long-term): Moderate signal, build over time

Impact: +50-60 minutes saved, focus on high-value prospects

---

Template-Driven Research

Pre-Built Templates

• Prospect dossier structure
• Verification protocol checklist
• Decision-maker mapping framework
• Value proposition calculator

Consistent Patterns

• Standard data collection (company, fraud signals, fit analysis)
• Reusable source verification (15 high-signal sources)
• Automated prioritization scoring

Impact: +30-40 minutes saved per task

---

Communication Protocol

Early Acceleration Detection

Signs of potential speedup:

• First 3-5 prospects completed faster than expected
• Parallel execution running smoothly
• No verification roadblocks
• High-signal sources yielding quick results

Action:

• Send progress update: "Accelerating faster than expected"
• Adjust ETA: "Completing ~3 hours early"
• Maintain quality standards

Real-Time Progress Updates

Instead of: Silent execution until completion Use:

• Hourly status (if task >2 hours)
• Early acceleration alerts
• Mid-task ETA adjustments

Impact: Reduced stakeholder anxiety, better expectations management

---

Token Efficiency Benchmarks

Target Metrics

• Tokens/prospect: 10-15k (sweet spot for quality)
• Output ratio: 3-5% of total tokens
• Token/hour: 300-400k (sustainable pace)

Red Flags

• 20k tokens/prospect = over-researching

• <8k tokens/prospect = potentially skipping verification
• <3% output ratio = excessive reasoning

• 500k tokens/hour = burning through efficiency

---

Quality Gates (Must Maintain)

Non-Negotiables

• ✅ All fraud signals verified from 2+ sources
• ✅ Decision-makers mapped (LinkedIn, org charts, SEC filings)
• ✅ Clear Arkose Labs value proposition for each prospect
• ✅ Recent incidents prioritized (last 12-18 months)
• ✅ Urgency signals flagged

Acceptable Trade-offs (if accelerating)

• ⚠️ Some Tier 3 prospects may have older incident data
• ⚠️ Decision-maker names may vary (role identification acceptable)
• ⚠️ Dark web claims flagged as [NEEDS VERIFICATION]

---

Implementation Checklist

Before Starting Task:

• Create prospect dossier templates
• Identify 15 high-signal data sources
• Prepare filtering criteria
• Set up parallel execution plan
• Establish communication protocol

During Task:

• Monitor execution speed in Hour 1
• Identify acceleration opportunities in Hour 2
• Send progress update if accelerating
• Document patterns for future tasks
• Maintain quality gates

After Task:

• Calculate actual runtime vs. estimate
• Document acceleration factors
• Update benchmarks if needed
• Share results with stakeholders
• Refine templates for next task

---

Success Metrics

Excellent Performance (9-10/10)

• 3x+ faster than estimate
• Zero quality compromises
• All deliverables complete
• High token efficiency

Good Performance (7-8/10)

• 2x+ faster than estimate
• Minor quality trade-offs acceptable
• All core deliverables complete
• Reasonable token usage

Needs Improvement (below 7/10)

• Slower than estimate
• Quality compromises
• Missing deliverables
• Inefficient token usage

---

Last Updated: 2026-04-11
Based on: 1 optimized research task (Arkose Labs)
Next Review: After 10 tasks (update benchmarks)