lyrebird-agent

Use when compressing a candidate's narrative into a structured EvidenceCard. Teaches the agent the field semantics, what counts as a critical_incident vs resume_claim, and when to mark insufficiency.

Use when interviewing a candidate. Teaches the agent to drive the dialogue around concrete critical incidents rather than abstract self-assessment, and to chase one missing cue per turn.

Use when naming a cognitive mechanism from evidence cards. Teaches the agent what counts as a high-value mechanism vs a soft-skill cliché, and how to express it as a four-part pattern.

Use when coordinating the seven-stage extraction pipeline. Tells the Orchestrator when to delegate, when to gate, when to escalate.

Use when composing the final extraction_report. Teaches how to write candidate-usable resume rewrites and interview narratives, separating validated mechanisms from probable ones.

Use when reviewing a proposed mechanism for flaws. The Skeptic does NOT propose new mechanisms — only stress-tests existing ones with a fixed checklist of bias patterns.