JD hard-requirement matcher
Use this skill to compare structured requirements with candidate or entity evidence. The skill is intentionally platform-neutral: it does not assume a browser, database, CRM, ATS, or messaging service.
Inputs
Accept either structured JSON or an equivalent table containing:
subject: the candidate/entity being evaluated;requirements: explicit hard requirements, each with an id, text, and optional evidence_policy;evidence: direct observations or source excerpts, each tied to a requirement when possible;- optional
context: job title, source, timestamp, and record identifiers.
Only use requirements supplied by the user or the calling workflow. Use lawful, job-relevant criteria and do not infer protected characteristics or invent thresholds.
Decision rules
- Evaluate every hard requirement independently.
- Direct evidence must support the requirement; keyword overlap alone is not evidence.
- Missing, ambiguous, negated, stale, or contradictory evidence does not satisfy a hard requirement.
- Stop deep analysis after a clearly failed hard requirement, but retain the reason and evidence used.
- Do not compensate for a failed hard requirement with unrelated strengths.
- Keep the output concise and do not reproduce an entire resume or source record.
Output
Return a compact result with:
decision: PASS, FAIL, or NEEDS_REVIEW;requirement_results: one result per requirement with status, evidence, and reason;matched_strengths: only when relevant;risks_or_missing_evidence;subject_ref and context_ref when supplied.
PASS requires every hard requirement to pass. Use NEEDS_REVIEW only when the workflow explicitly allows human review for unresolved evidence; otherwise treat unresolved evidence as FAIL.
Boundaries
- Do not contact, message, forward, reject, or mutate the source system.
- Do not persist data unless the calling workflow explicitly provides a storage contract.
- Do not expose sensitive source text beyond the minimum evidence needed for the decision.
1---2name: jd-hard-requirement-matcher3description: Evaluate a candidate or record against explicit, lawful, job-relevant hard requirements and return an evidence-based pass/fail decision. Use when a workflow needs deterministic requirement matching without relying on keyword similarity or hidden criteria.4---56# JD hard-requirement matcher78Use this skill to compare structured requirements with candidate or entity evidence. The skill is intentionally platform-neutral: it does not assume a browser, database, CRM, ATS, or messaging service.910## Inputs1112Accept either structured JSON or an equivalent table containing:1314- `subject`: the candidate/entity being evaluated;15- `requirements`: explicit hard requirements, each with an `id`, `text`, and optional `evidence_policy`;16- `evidence`: direct observations or source excerpts, each tied to a requirement when possible;17- optional `context`: job title, source, timestamp, and record identifiers.1819Only use requirements supplied by the user or the calling workflow. Use lawful, job-relevant criteria and do not infer protected characteristics or invent thresholds.2021## Decision rules22231. Evaluate every hard requirement independently.242. Direct evidence must support the requirement; keyword overlap alone is not evidence.253. Missing, ambiguous, negated, stale, or contradictory evidence does not satisfy a hard requirement.264. Stop deep analysis after a clearly failed hard requirement, but retain the reason and evidence used.275. Do not compensate for a failed hard requirement with unrelated strengths.286. Keep the output concise and do not reproduce an entire resume or source record.2930## Output3132Return a compact result with:3334- `decision`: `PASS`, `FAIL`, or `NEEDS_REVIEW`;35- `requirement_results`: one result per requirement with `status`, `evidence`, and `reason`;36- `matched_strengths`: only when relevant;37- `risks_or_missing_evidence`;38- `subject_ref` and `context_ref` when supplied.3940`PASS` requires every hard requirement to pass. Use `NEEDS_REVIEW` only when the workflow explicitly allows human review for unresolved evidence; otherwise treat unresolved evidence as `FAIL`.4142## Boundaries4344- Do not contact, message, forward, reject, or mutate the source system.45- Do not persist data unless the calling workflow explicitly provides a storage contract.46- Do not expose sensitive source text beyond the minimum evidence needed for the decision.
lbql-yhl