Foresight

Foresight

Estimate likely outcomes before committing to a plan, change, or launch.

What This Skill Does

Foresight produces bounded forecasts with explicit uncertainty to guide decisions. It is NOT:

• Preflight checks: Verifying prerequisites before executing a known action (disk space, backups, connection strings). Those are safety gates, not forecasts.
• Diagnosis/Debugging: Finding what's broken right now (null pointers, deprecated APIs). Those are root-cause analyses, not predictions.
• Monitoring: Watching real-time metrics or alerting on thresholds. Those are observability tasks, not forward-looking estimates.
• Reconnaissance: Gathering facts about competitors or systems. Intelligence gathering feeds foresight but isn't foresight itself. Key distinction: If the user wants to know "what will likely happen if we choose X," use Foresight. If they want to know "is it safe to run X now," "what's broken," or "what are they doing," use a different spell. In this grimoire, Foresight is treated as a metaphorical spell with a shipping-now delivery profile. Canonical reference input: Foresight (spell).

When To Use

• Activate this spell when the user asks for a forward-looking, probability-weighted forecast to inform a decision. Look for:
• Explicit choice between options ("should we X or Y?", "migrate vs stay", "build vs buy")
• Time-bounded outcome requests ("over the next 12 months", "by Q3", "18-month trajectory")
• Risk/uncertainty language ("risk-weighted", "confidence level", "probability", "best case/worst case", "weal or woe")
• Decision frameworks with explicit unknowns ("what would change your recommendation?", "decisive unknowns")

Prerequisites

• No extra runtime dependencies beyond Hermes Agent and the normal toolset for this session.

Procedure

1. Restate the target, the success condition, and any no-touch boundaries before taking action.
2. Scope the decision: Restate the choice, the time horizon, and what success/failure looks like.
3. Map the paths: List the strongest upside and downside scenarios for each option.
4. Estimate from evidence: Assign likelihoods based on current data, not vibes. State your assumptions explicitly.
5. Return the forecast: Deliver (a) a recommendation with confidence level, (b) a short risk matrix, and (c) the decisive unknowns that would change the call.
6. Package the result as the deliverables below, with confidence, assumptions, and unresolved risk called out explicitly.

Deliverables

• Recommendation: [clear choice] (Confidence: X%) Assumptions: [what you're basing this on]
• | Path | Upside | Downside | Likelihood |
• Decisive Unknowns: [1-3 things that would change this call] ```

Pitfalls / Guardrails

• Keep the metaphor anchored to a real mechanism instead of drifting into lore.
• Do not use for: "Check if the backup completed before I run this" → Preflight
• Do not use for: "Find all deprecated API calls in our codebase" → Scan/Diagnosis
• Do not use for: "Alert me if error rates exceed 2% this hour" → Monitoring
• Do not use for: "What features is our competitor shipping?" → Reconnaissance
• Do not use for: "Why is this function throwing a null pointer?" → Diagnosis
• Do not use for: Activate only when the core request is forecasting outcomes to inform a choice under uncertainty.

Verification

• Check that the result includes every deliverable promised above.
• Check that confirmed facts, assumptions, and inferences are visibly separated.
• Check that the metaphor still maps cleanly to a real operational mechanism.

Example Invocation

/foresight evaluate this decision, give me the likely upside and downside, and tell me what would change your call