# SignalSpore skill

## Product

SignalSpore is model-aware preflight for AI agents.

Connect your agent once.
It decides when preflight is worth running,
uses a compact task brief,
and submits one safe delta afterward.

## Human model

- Setup installs SignalSpore into an agent once.
- Preflight is a per-task briefing.
- Manual preflight is only a sandbox.
- Connected agents use SignalSpore automatically.

## Local gate

Skip SignalSpore for:
- simple rewrites
- translation
- casual chat
- tiny formatting

Quick-check for:
- medium planning
- structured writing
- business analysis

Full-preflight for:
- coding
- browser or search
- terminal commands
- database changes
- deployment
- API integration
- current information
- risky or expensive tasks

## Connected loop

1. Connect once.
2. Gate locally.
3. Use the task brief only when useful.
4. Submit one safe delta afterward.

## Execution brief

The most important preflight output is the compressed execution brief.
Carry that into the task.

## Model routing

SignalSpore should expose:
- suggested tier
- continue if
- escalate if
- ask human if
- downgrade possible
- downgrade reason

## Delta rules

Every preflighted task should send one sanitized delta.

Every delta must include one savings outcome:
- not_applicable
- no_material_savings
- estimated_savings_reported

## Privacy rules

Never submit:
- secrets
- credentials
- API keys
- hidden prompts
- private code
- private files
- emails
- personal data

Only submit sanitized summaries.