Reports

Sills reports separate observed evidence, automated results, inference, and manual-review requirements. Use these pages to read the generated files and decide what another agent should do next.

For concrete output shape and language, read the example reports.

Report structure

A full report contains navigation, summary, Markdown and JSON reports, a manifest, coverage, specialist reports, finding files, evidence, raw outputs, and a remediation handoff.

Findings

Each finding has a stable ID, kind, category, severity, confidence, status, origin, scope, impact, evidence, recommendation, and verification steps.

Evidence

Evidence is referenced from findings and indexed in the manifest. Sensitive data must be redacted. Inference is never presented as direct observation.

Agent handoff

Tell another agent to read handoff.md and report.json, resolve open findings in priority order, preserve positive patterns, verify every change, and retain original evidence.

Severity and release blockers

Sills uses critical, major, moderate, minor, and observation severity. Severity describes impact; release-blocker status separately describes whether the current release should stop.

Release decisions consider task blockage, affected users, frequency, recoverability, legal or compliance context, data and financial risk, security exposure, systemic reach, and workarounds.

Confidence

Confirmed, high, medium, low, and requires-manual-verification describe evidence strength. Confidence never replaces severity.

Direct runtime reproduction is different from source evidence, automated detection, documentation, or inference. Reports preserve that distinction for every finding.