extract-evidence

Renamed: extract-evidence was build-candidate-evidence before v1.0.0 (see RIT-A-0005).

Inputs must be canonical JSON. This capability consumes a resume as a ResumeDocument JSON (build it from a PDF/DOCX/MD/text file with the parse-resume skill) and, where a job is involved, a JobDescription JSON (build it with parse-job so skills-coverage scoring works). Run those conversions in subagents, then pass the saved JSON paths here — they live under resume-kit/resumes/ and resume-kit/jobs/.

Prerequisites

Run the shared Prerequisites gate first — see _shared/prerequisites.md.

• Required input: a resume/master ResumeDocument JSON (config.json active_resume or an explicit path).
• If missing: STOP and run parse-resume to produce the ResumeDocument JSON before calling this skill.

Purpose

Produce structured CandidateEvidence records that ground subsequent update-keywords / update-terminology and validate-facts calls. Evidence records capture what the candidate can truthfully claim, preventing the alignment engine from fabricating experience.

When to use

• Before running update-keywords or update-terminology with evidence constraints.
• Before running validate-facts to produce an evidence baseline.
• When the user wants to review and approve their own claims before alignment.

Inputs

Field	Type	Notes
resume	ResumeDocument	Source resume from which evidence is derived
approved_claims	list[CandidateEvidence] | list[str] | null	Optional pre-approved claims to include (bypasses engine derivation for those items)

Options (CapabilityOptions)

Option	Default	Effect
strict	false	Escalate advisory warnings to failures

no_llm has no effect — this capability is always deterministic.

How to invoke

CLI

resume-tool build-evidence --resume <resume.json> \
    [--approved-claims <claims.json>] \
    [--output {json,text,md}] [--strict]