ee-skill-standards

What problem does it solve?

It prevents project-local ee skills from becoming un-auditable, unsafe, or overly speculative by enforcing an evidence-bound boundary between mechanical ee JSON outputs and agent-judgment behavior.

Core Features & Use Cases

• Mechanical Command Boundary: Skills interpret stdout JSON from explicit ee commands and avoid parsing progress, human text, or traces as evidence.
• Evidence Bundle-First Handoff: Skills should prefer ee.skill_evidence_bundle.v1 (optionally also ee.response.v1 / ee.error.v1) and use provenance and hashes instead of direct DB/index access.
• Stop/Go Gates and Uncertainty Handling: Skills must stop on missing/invalid JSON, unknown redaction status, degraded invalidation, stale/mismatched bundles, or prompt-injection-like unquarantined data.
• Explicit Output Template for Decisions: Standardized sections (Decision, Evidence, Uncertainty, Degraded State, Recommended Explicit Commands) make judgments easier to review and audit.
• Testing and E2E Logging Requirements: Skills must include lint-like static checks plus degraded/refusal path coverage and consistent logging for reproducibility.

Quick Start

Use the ee-skill-standards when creating or updating a project-local skill that consumes mechanical ee CLI JSON and must apply agent judgment with strict evidence handling and boundaries.