技能列表

Compare baseline and candidate agent metrics and create a blocking remediation queue for regressions.

Compare observed agent tool calls with expected tools, argument requirements, and forbidden operations.

Normalize redacted multi-agent and tool spans into deterministic local observability records.

Validate ownership, classification, visibility, effective time, and expiration metadata before indexing.

Compare previous and current knowledge sources and produce create, update, delete, and unchanged queues.

Create deterministic text chunks with overlap, hashes, stable IDs, and short-chunk quality warnings.

Compare desired chunks with an existing vector-index manifest and produce reviewed upsert, delete, and rollback queues.

Use when planned AI-agent tool actions must be routed to allow, human review, or block based on side effects, personal data, and rollback support.

Use when a team needs an auditable inventory of AI-agent SaaS connectors, owners, environments, approval state, and risky production write access.

Use when approved SaaS connector actions need a tamper-evident audit chain, replay order, failure evidence, and a human-reviewed rollback queue.

Use when an AI-agent integration needs to compare granted OAuth scopes with the minimum scopes required by approved business operations.

Use when a voice-agent team needs per-turn cost accounting, budget stops, consent checks, privacy review, latency checks, and an auditable human-review queue.