SkillPulse

Detect, score, and heal broken agent skills.

SkillPulse is the reliability layer for agent skills.

When models, tools, APIs, prompts, or codebases change, agent skills can silently degrade. SkillPulse helps detect breakage early, score skill health, run regression checks, and route fallbacks before failures cascade into production incidents.

Why SkillPulse

Agent skills are often treated as static assets, but the environments around them are not static.

Skills degrade because:

• models change
• tool signatures change
• APIs evolve
• prompts drift
• dependencies break
• runtime assumptions become invalid
• task distributions shift over time

SkillPulse exists to make skill reliability observable and actionable.

Core goals

• Parse and validate skill manifests
• Detect compatibility issues
• Score skill health
• Run lightweight regression checks
• Surface degradation early
• Suggest fallback or quarantine actions

Initial scope

SkillPulse v0 starts with:

• skill.yaml manifest parsing
• health status classification
• basic compatibility checks
• health scoring
• CLI inspection commands

Later phases may add:

• regression suites
• fallback routing
• quarantine policies
• event hooks
• self-healing suggestions
• framework adapters

Example

skillpulse check ./examples/researcher-web

Example output:

Skill: researcher-web Status: degraded Score: 72/100

Detected issues:

• required tool signature mismatch
• missing fallback model
• latency threshold exceeded

Philosophy

SkillPulse is not another agent framework.

It is a portable reliability layer that can sit on top of existing agent runtimes and broader agent governance architectures.

Roadmap

See:

• docs/vision.md
• docs/architecture.md
• docs/roadmap.md skillpulse score ./examples/researcher-web skillpulse doctor ./examples/researcher-web